# Changelog

All notable changes to **Pipecat** will be documented in this file.

The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.0.0/),
and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).

<!-- towncrier release notes start -->

## [1.12.0] - 2026-09-25

### Added

- TTS sentence aggregation follows `Settings.language`, defaulting to English
  when unspecified. Language updates apply with TTS settings.
  (PR [#5737](https://github.com/pipecat-ai/pipecat/pull/5737))

- Added `TTSService.pronunciation_transform_ipa()`, which builds a text
  transform from a word-to-IPA mapping so a voice says names and terms it would
  otherwise guess from spelling. Each matched word is replaced with the
  service's `format_pronunciation()` output, so one mapping works with any
  service that supports pronunciation hints; a word the service cannot use is
  reported once when the transform is built and spoken as written. Cartesia
  writes IPA as inline phoneme blocks, ElevenLabs as SSML `<phoneme>` tags
  (read by `eleven_flash_v2` and `eleven_turbo_v2`, and on the WebSocket
  service only with `enable_ssml_parsing=True`; on any other setup the
  transform is skipped with a warning and words are spoken as written), Eleven
  v3 and Inworld as IPA between slashes, and Deepgram Aura-2 as an inline
  pronunciation object. Deepgram Flux has no pronunciation markup, so its words
  are spoken as written. `pipecat.utils.text.phonemes` holds the IPA parsing
  and normalization the formatters share. Register the transform last in
  `text_transforms`, so no later transform rewrites the markup it inserts.

    ```python
    pronounce = CartesiaTTSService.pronunciation_transform_ipa({"Metformin":
  "mɛtˈfɔɹmɪn"})

    tts = CartesiaTTSService(
        text_transforms=[
            ("*", strip_markdown),
            ("*", pronounce),  # last, so nothing rewrites its markup
        ],
    )
    ```
  (PR [#5798](https://github.com/pipecat-ai/pipecat/pull/5798))

- Added client-mode reconnect to `MOQTransport`: when the relay session drops,
  the transport redials with backoff for as long as
  `MOQParams.connection_timeout` allows the client to be missing, keeps its
  broadcast across dials, and reports the client gone only when that time is
  up. A relay that vanishes without closing the session is detected by watching
  the connection's traffic counters, rather than waiting out the QUIC idle
  timeout. Each side now sends a `session-ending` marker on its transcript
  before leaving; a peer whose tracks end without it is treated as an outage
  rather than a hangup. `on_disconnected` and `on_connected` now fire for each
  relay session, so a redial shows up as a disconnect followed by a connect;
  end the call from `on_client_disconnected`, which fires only once the client
  is gone.
  (PR [#5800](https://github.com/pipecat-ai/pipecat/pull/5800))

- Added `MOQRunnerArguments.relay_url` for dialing a relay by its full URL,
  query string included. `host` and `port` are optional when it is set, and
  `create_transport` passes it through to `MOQParams.relay_url`. When `host` or
  `port` is given as well, `relay_url` wins and a warning is logged.
  (PR [#5800](https://github.com/pipecat-ai/pipecat/pull/5800))

- Added `interruptible` to every frame: True by default and False by default
  for `UninterruptibleFrame` subclasses, and what the frame queue, the
  processor's interruption handling and the speculation gate decide by. Set it
  on a frame before pushing it to keep that one frame through an interruption,
  or to let one frame of a protected type be dropped, without a frame class of
  your own.
  (PR [#5835](https://github.com/pipecat-ai/pipecat/pull/5835))

- `JudgeVerdict` now has a `confidence`, from 0 to 1, saying how sure the judge
  is. With Jev the number is calibrated. With an LLM it is the LLM's own guess.
  (PR [#5857](https://github.com/pipecat-ai/pipecat/pull/5857))

- The eval judge can now use a classifier. Put a `factory:` in the scenario's
  `judge.eval:` block that returns a `BaseClassifier`, or an LLM service as
  before. Our release evals judge with Jev this way, through `evals/judges.py`.
  Jev answers each check in a few hundred milliseconds and says how sure it is.
  It needs the `jev` extra and `TYPESAFE_API_KEY`.
  (PR [#5857](https://github.com/pipecat-ai/pipecat/pull/5857))

- Added `allow_continue` to a scenario's `judge.eval:` block and to
  `EvalJudge`. Set it to `false` when every reply you judge is a final answer.
  The judge then answers yes or no and never `continue`, so it never waits for
  more text that isn't coming.
  (PR [#5857](https://github.com/pipecat-ai/pipecat/pull/5857))

- Added `LLMClassifier`, which answers classifier questions with any Pipecat
  LLM service that supports `run_inference()`. All the questions about one
  state go to the LLM in one call, and the LLM replies with one JSON object
  holding an answer per question. It waits `timeout` seconds for the LLM, 10 by
  default, and raises `ClassifierError` if the LLM does not answer in time or
  the call fails.

    ```python
    classifier = LLMClassifier(llm=OpenAILLMService(model="gpt-4o-mini"))
    results = await classifier.yes_no(
        "Hi, you've reached Dana. Leave a message.",
        {"voicemail": YesNoQuestion(instructions="is this a voicemail greeting?")},
    )
    results["voicemail"].is_yes  # True
    results["voicemail"].probability  # 0.97
    ```
  (PR [#5863](https://github.com/pipecat-ai/pipecat/pull/5863))

- Classifiers report metrics through an `on_metrics` event: after every call,
  the time it took as `ProcessingMetricsData` and, for `JevClassifier`, the
  tokens it used as `LLMUsageMetricsData`. A classifier cannot push frames, so
  its owner puts the data in a `MetricsFrame`.
  (PR [#5863](https://github.com/pipecat-ai/pipecat/pull/5863))

- Added Pipecat Classifiers. A classifier is a small object that answers typed
  questions about some state. A question is a `YesNoQuestion`, a
  `ChoiceQuestion` among options, or a `ScoreQuestion` on a scale. Questions
  are asked by name, several about one state at once. `ask()` takes any mix of
  kinds, and `yes_no()`, `choice()` and `score()` take questions of one kind
  and return typed results: a `YesNoResult` with the probability of yes and
  `is_yes`, a `ChoiceResult` with the `choice` and a probability per option,
  and a `ScoreResult` with the `score` on the scale and a probability per
  level. `JevClassifier` answers them through Jev, TypeSafe's classification
  model, in one request. A choice question can have at most 255 options, Jev's
  limit (`JEV_MAX_CHOICE_OPTIONS`). Several classifiers can share one
  `JevClient`, which handles the HTTP/2 connection, auth, retries and token
  accounting. Install with `uv add "pipecat-ai[jev]"`.
  `examples/features/features-classifiers.py` asks one question of each kind.

    ```python
    classifier = JevClassifier(api_key=os.getenv("TYPESAFE_API_KEY"))
    results = await classifier.choice(
        "I'd like to book a table for, um",
        {
            "turn": ChoiceQuestion(
                instructions="is the user's turn over?",
                options={
                    "complete": "the user finished",
                    "short": "a brief pause",
                    "long": "asked for time",
                },
            )
        },
    )
    results["turn"].choice  # "short"
    ```
  (PR [#5863](https://github.com/pipecat-ai/pipecat/pull/5863))

- Added a read-only `settings` property to `AIService`, so code outside a
  service can read its current settings, such as the model. Settings are still
  changed through a `ServiceUpdateSettingsFrame`.
  (PR [#5863](https://github.com/pipecat-ai/pipecat/pull/5863))

- Added an optional `response_schema` argument to `LLMService.run_inference()`,
  a JSON schema the reply must follow. OpenAI, Anthropic and Google have the
  provider enforce it, so the reply is JSON text in that shape. A service or
  model that cannot enforce one, such as Bedrock, DeepSeek or an OpenAI model
  before gpt-4o-mini, ignores it with a warning, and `supports_response_schema`
  says whether a service can at all.

    ```python
    reply = await llm.run_inference(context, response_schema=schema)
    ```
  (PR [#5876](https://github.com/pipecat-ai/pipecat/pull/5876))

- Added `LLMUserAggregatorParams.empty_user_turn`, an `EmptyUserTurnConfig` for
  user turns that end with no transcript, such as a cough, background noise, or
  speech the STT could not recognize. It is on by default, even when
  `empty_user_turn` isn't passed, and treats two cases differently:

    - Interrupted: the turn cut the bot off. The bot would otherwise stay silent
      mid-response, so the aggregator appends a developer message
      (`interrupted_prompt`) and runs the LLM once, and the bot asks the user to
      repeat or picks up where it left off. On by default.
    - Idle: the bot had finished and was waiting for the user. The conversation
      isn't stuck, and answering what may be noise would be intrusive, so these
      turns get no answer unless `idle_prompt` is set.

    ```python
    LLMUserAggregatorParams(
        empty_user_turn=EmptyUserTurnConfig(
            idle_prompt=(
                "The user may have said something, but it was not "
                "recognized. Briefly ask them to repeat it."
            ),
        ),
    )
    ```

  Setting `interrupted_prompt=None` turns off the interrupted case, and
  `empty_user_turn=None` turns off both. `max_consecutive_recoveries` (default
  1) limits how many empty turns in a row get an answer.
  (PR [#5907](https://github.com/pipecat-ai/pipecat/pull/5907))

- `UIWorker` can now use a classifier for small decisions about the screen,
  with no LLM turn: which element the user means (`which_element`), whether
  something is true on the screen (`check_screen`), which elements match a
  description (`select_elements`), whether a UI event deserves a reply
  (`should_respond`), and `act` on an element named in words. Pass a
  `classifier`, for example a `JevClassifier`. If you don't, the worker's own
  LLM answers through an `LLMClassifier`, which works but is slower.

    ```python
    worker = MyUIWorker("ui", llm=OpenAILLMService(api_key=...),
  classifier=JevClassifier(api_key=...))
    ref = await worker.which_element("the checkout button")
    ```
  (PR [#5909](https://github.com/pipecat-ai/pipecat/pull/5909))

- A voice LLM can now ask a `UIWorker` about the screen without ever seeing it.
  The worker has a built-in `screen` job with an `action`: `find` (which
  element a description means), `check` (whether something is true on the
  screen), `select` (which elements match a description), `list` (what is on
  the screen), `selection` (the text the user has selected) and `click`,
  `scroll_to`, `highlight`, `select_text` or `fill` (do that to the element a
  description means). Every answer is short data, never the page.
  `screen_tools("ui")` gives the voice LLM the tool that sends it.

    ```python
    context = LLMContext(tools=screen_tools("ui"))
    ```
  (PR [#5909](https://github.com/pipecat-ai/pipecat/pull/5909))

- `UIWorker` now has a `snapshot` property with the latest page snapshot and a
  `selection` property with the text the user has selected, so a subclass can
  read the screen with plain code.
  (PR [#5909](https://github.com/pipecat-ai/pipecat/pull/5909))

- Added `RTVIFunctionCallReportLevel.ARGUMENTS`, between `NAME` and `FULL`. It
  reports the function name and the arguments, and no result.
  (PR [#5918](https://github.com/pipecat-ai/pipecat/pull/5918))

### Changed

- `InworldRealtimeLLMService` now sends
  `providerData.auto_tool_response=false`, leaving Pipecat responsible for
  requesting continuation after submitting a tool result.
  - `InworldRealtimeLLMService` now generates a unique session key for every
  connection, including connections opened within the same millisecond.
  (PR [#5116](https://github.com/pipecat-ai/pipecat/pull/5116))

- Sentence aggregation uses self-contained sentencex rules without NLTK data
  downloads or tokenizer warm-up. Language-specific sentence boundaries may
  differ from Punkt.
  (PR [#5737](https://github.com/pipecat-ai/pipecat/pull/5737))

- `MOQTransport` numbers every transcript record with `seq` and `epoch` and
  drops replayed records on subscribe, so a reconnect on either side no longer
  redelivers the whole RTVI log and re-fires `client-ready`. Both fields are
  stripped before the message reaches the pipeline; records without them pass
  through unchanged, so older peers keep working.
  (PR [#5800](https://github.com/pipecat-ai/pipecat/pull/5800))

- `MOQTransport` failures now reach the pipeline as an `ErrorFrame` with an
  error category, in addition to the `on_error` event. A relay that refuses the
  token, at the dial or by closing the session as unauthorized, is not retried
  and marks the transport unusable, as does a relay that cannot be reached
  within `connection_timeout`.
  (PR [#5800](https://github.com/pipecat-ai/pipecat/pull/5800))

- The `moq` extra requires `moq-rs` 0.4.6 or later. `moq-rs` is the client
  library `MOQTransport` dials a relay with; `moq-relay` is the separate
  server. The transport is tested against `moq-relay` 0.14. A relay from an
  incompatible release line can accept the connection and still drop the
  transcript track without an error.
  (PR [#5800](https://github.com/pipecat-ai/pipecat/pull/5800))

- `MOQParams.connection_timeout` is now the one limit on how long the client
  may be missing, and defaults to 60 s (was 30 s). It bounds the wait for the
  client to join and, in client mode, an outage: the time from the relay
  session dropping until the client's data flows again, across every redial.
  Set it longer than the time your load balancer takes to fail a dead relay
  out.
  (PR [#5800](https://github.com/pipecat-ai/pipecat/pull/5800))

- `XAISTTService` now sends its `model` setting to xAI, and defaults to
  `grok-voice-transcribe-2.0`. Previously no model was sent, so xAI used its
  server default, `grok-voice-transcribe-1.0`. To keep the previous model, pass
  `settings=XAISTTService.Settings(model="grok-voice-transcribe-1.0")`.
  (PR [#5847](https://github.com/pipecat-ai/pipecat/pull/5847))

- The `local-smart-turn` and `moondream` extras now require
  `transformers>=5.10.0`. The `moondream` extra's `accelerate`, `einops`,
  `pyvips` and `timm` pins are relaxed to a lower bound with a major-version
  cap.
  (PR [#5850](https://github.com/pipecat-ai/pipecat/pull/5850))

- `EvalJudge` now decides every verdict with a classifier. By default that is
  an `LLMClassifier` over the LLM the `judge.eval:` block names, so existing
  scenarios work as before. A classifier gives no reasons, so the judge asks an
  LLM, the explainer, for the reason behind every `no` and every verdict below
  `explain_below`. The explainer is the judging LLM unless an `explainer:`
  block names another one, and `explainer: false` turns reasons off. The
  classifier's verdict always stands.
  (PR [#5857](https://github.com/pipecat-ai/pipecat/pull/5857))

- `pipecat eval suite` now keeps `concurrency` runs going at all times, taking
  the next run in manifest order. Before, each entry ran its scenarios one at a
  time on one slot, so a manifest with fewer entries than slots left slots
  idle, and one entry with ten scenarios ran them one by one. An entry whose
  provider rate-limits sets its own `concurrency:` to cap its runs in flight,
  as the turn-completion manifest does.
  (PR [#5857](https://github.com/pipecat-ai/pipecat/pull/5857))

- The `mcp` extra now requires `mcp` 2 (`mcp[cli]>=2.1.1,<3`). `MCPClient`
  works the same; it no longer runs on the 1.x SDK.
  (PR [#5857](https://github.com/pipecat-ai/pipecat/pull/5857))

- An LLM judge now classifies instead of answering a prose prompt: it gets the
  conversation as structured state and picks one of the outcomes. A borderline
  reply can get a different verdict than before. A simulation is judged one bot
  turn per call instead of the whole run in one call, so it costs more calls.
  (PR [#5857](https://github.com/pipecat-ai/pipecat/pull/5857))

- `VoicemailDetector` now takes a `classifier` and is a single processor
  instead of a parallel pipeline with its own LLM. It asks the classifier after
  each transcription and acts once the caller has been quiet for
  `decision_timeout` (default 1 s), when the latest answer decides. No verdict
  acts on a fragment: "hi, this is Sam" is what a person says and how a
  greeting starts, and only the silence that follows tells them apart. Then the
  held-back speech is released or dropped as before. The
  `on_voicemail_detected` and `on_conversation_detected` handlers receive the
  detector itself.

    ```python
    # Before
    detector = VoicemailDetector(llm=OpenAILLMService(api_key=...))

    # After
    detector = VoicemailDetector(classifier=JevClassifier(api_key=...))
    ```
  (PR [#5869](https://github.com/pipecat-ai/pipecat/pull/5869))

- Observers created with `observe_every_push=False` are told about a frame
  once, on its first push, instead of on every push by every processor that
  passes it along. The built-in observers that handle a frame once do so, and
  `FramePushed.first_push` tells the first push from the ones that follow for
  the ones that observe every push.
  (PR [#5908](https://github.com/pipecat-ai/pipecat/pull/5908))

- A `UIWorker` now answers a `respond` job with the reply its LLM writes, so
  the smallest UI worker is a `UIWorker` with an LLM and a system prompt. A
  `@tool` that calls `respond_to_job` still answers instead when it needs to,
  for example to speak the answer through TTS.
  (PR [#5909](https://github.com/pipecat-ai/pipecat/pull/5909))

- `UI_SNAPSHOT_EVENT_NAME` and `UI_CANCEL_JOB_GROUP_EVENT_NAME` are now public
  in `pipecat.bus.ui`. They are the bus event names of the client's screen
  snapshot and of the client asking to cancel a job group.
  (PR [#5909](https://github.com/pipecat-ai/pipecat/pull/5909))

### Deprecated

- Deprecated `UninterruptibleFrame`. A frame class that should be
  uninterruptible by default declares `interruptible: bool =
  field(default=False, init=False)` instead. The marker still sets the flag
  until it is removed in 2.0.0.
  (PR [#5835](https://github.com/pipecat-ai/pipecat/pull/5835))

- Passing an LLM service to `EvalJudge`, as the first argument or as
  `service=`, is deprecated and will be removed in 2.0.0. Pass
  `EvalJudge(LLMClassifier(llm=service), explainer=service)` instead, which is
  what the old form did.
  (PR [#5857](https://github.com/pipecat-ai/pipecat/pull/5857))

- Deprecated the startup warming timings reported by `StartupTimingObserver`,
  all removed in 2.0.0: the `StartupTimingReport.warmup` field,
  `StartupWarmupTiming`, `StartupWarmup`, and the `on_startup_warmup()` hook on
  `BaseObserver`, `WorkerObserver` and `StartupTimingObserver`. Pipecat warms
  no deferred imports while the pipeline sets up, so `warmup` is always `None`
  and the hook is never called. The rest of the report — the phase totals and
  the per-processor timings — is unchanged.
  (PR [#5859](https://github.com/pipecat-ai/pipecat/pull/5859))

- `VoicemailDetector`'s `llm` and `custom_system_prompt` parameters are
  deprecated. Pass a `classifier` instead. Until removal, the `llm` is wrapped
  in an `LLMClassifier`, with the custom prompt in front of the classifier's
  instructions.
  (PR [#5869](https://github.com/pipecat-ai/pipecat/pull/5869))

- The `max_frames` parameters of `TurnTrackingObserver` and
  `UserBotLatencyObserver` are now deprecated, removed in 2.0.0. Observers no
  longer keep a window of the frames they have seen.
  (PR [#5908](https://github.com/pipecat-ai/pipecat/pull/5908))

- `tts_speak` on `UIWorker.respond_to_job` is deprecated and will be removed in
  2.0.0. A UI worker should not speak; respond with the answer and let the
  voice LLM say it.
  (PR [#5909](https://github.com/pipecat-ai/pipecat/pull/5909))

- `BaseUIWorker` is deprecated and will be removed in 2.0.0. Use `UIWorker`
  instead. `UIWorker` now reports its job groups to the client on its own, so
  instead of a separate `BaseUIWorker` dispatcher, dispatch job groups from a
  `@job` handler in your `UIWorker`.
  (PR [#5909](https://github.com/pipecat-ai/pipecat/pull/5909))

- `ReplyToolMixin` is deprecated and will be removed in 2.0.0. Use
  `screen_tools` instead: the voice LLM asks the UI worker through the `screen`
  job and says the answer itself.
  (PR [#5909](https://github.com/pipecat-ai/pipecat/pull/5909))

### Removed

- Removed `NotifierGate`, `ClassifierGate`, `ConversationGate` and
  `ClassificationProcessor` from
  `pipecat.extensions.voicemail.voicemail_detector`, and the
  `CLASSIFIER_RESPONSE_INSTRUCTION` and `DEFAULT_SYSTEM_PROMPT` attributes of
  `VoicemailDetector`. They were parts of the old parallel-pipeline detector
  and its prompt.
  (PR [#5869](https://github.com/pipecat-ai/pipecat/pull/5869))

### Fixed

- Fixed an `LLMService` regression where re-advertising a tool with the same
  name but a different handler (for example, a per-node handler in Pipecat
  Flows) silently kept the previous handler bound. Auto-registered handlers are
  now rebound when the advertised handler changes, while explicit
  `register_function` registrations are still left untouched.  (PR
  [#4823](https://github.com/pipecat-ai/pipecat/pull/4823))

- Fixed `InworldRealtimeLLMService` resending a server-VAD user transcript
  after a fast tool result context update, which duplicated user turns and
  responses.
  (PR [#5116](https://github.com/pipecat-ai/pipecat/pull/5116))

- `InworldRealtimeLLMService` now serializes Inworld extensions under
  `providerData` instead of `provider_data`, allowing the server to apply them.
  (PR [#5116](https://github.com/pipecat-ai/pipecat/pull/5116))

- Applied NVIDIA STT settings changes to the running stream. `NvidiaSTTService`
  rebuilt its recognition config on a settings update but never reconnected, so
  the open gRPC stream kept transcribing with the previous settings while
  `self._settings` reported the new ones.
  (PR [#5632](https://github.com/pipecat-ai/pipecat/pull/5632))

- Fixed Flows node transitions where an interruption could leave the LLM
  running with the previous node's context and tools. The
  `LLMMessagesAppendFrame` or `LLMMessagesUpdateFrame` and the
  `LLMSetToolsFrame` a transition queues are now uninterruptible, so they are
  still delivered.
  (PR [#5837](https://github.com/pipecat-ai/pipecat/pull/5837))

- `TelnyxFrameSerializer` now sends `OutputTransportMessageFrame` and
  `OutputTransportMessageUrgentFrame` to the client as JSON, matching the other
  telephony serializers. Previously these messages (including RTVI messages)
  were silently dropped.
  (PR [#5841](https://github.com/pipecat-ai/pipecat/pull/5841))

- Fixed the eval harness treating the bot's earlier speech as its reply when a
  scenario interrupts the bot. Speech from before the interruption is now
  dropped.
  (PR [#5846](https://github.com/pipecat-ai/pipecat/pull/5846))

- Fixed `MoondreamService` with transformers 5: loading the model on a GPU or
  Apple Silicon failed with `AttributeError: 'HfMoondream' object has no
  attribute 'all_tied_weights_keys'`, and a model that did load produced
  garbage descriptions.
  (PR [#5856](https://github.com/pipecat-ai/pipecat/pull/5856))

- Fixed a bridged `PipelineWorker` reporting every frame from its LLM twice
  over RTVI, once in its own pipeline and once when the frame crossed the
  bridge, which doubled every word of a bridged worker's reply in the client's
  LLM text and in the evals. `enable_rtvi` now defaults to off for a bridged
  pipeline worker, which has no client of its own; `LLMWorker` already did
  this. Pass `enable_rtvi=True` to keep it.
  (PR [#5857](https://github.com/pipecat-ai/pipecat/pull/5857))

- Fixed TTS word tracking falling out of step on markdown-heavy replies. Most
  word-timestamp events were dropped with "Dropping word ... not recognised by
  any slot" warnings, and the rest of the reply arrived in large chunks, so
  word highlighting stalled and then jumped ahead. Two token shapes triggered
  it: the period Cartesia adds to the last token of a line (`images:.`,
  `---.`), and a symbol a provider reports differently from the text
  (ElevenLabs reports `→` as `-`) followed by more symbols such as `###` or
  `**`.
  (PR [#5866](https://github.com/pipecat-ai/pipecat/pull/5866))

- Fixed `WorkerRunner` skipping a worker that another worker added while the
  runner was still starting, for example a child worker added by a processor
  during its setup.
  (PR [#5872](https://github.com/pipecat-ai/pipecat/pull/5872))

- Fixed `pcm_to_wav()` dropping complete samples when given a typed
  `memoryview`.
  (PR [#5879](https://github.com/pipecat-ai/pipecat/pull/5879))

- Fixed `is_silence()` misclassifying full-scale negative int16 PCM samples as
  silence.
  (PR [#5881](https://github.com/pipecat-ai/pipecat/pull/5881))

- Fixed `MOQTransport` ending the call when a relay refused a subscription to
  the peer's broadcast with `dropped`, which a surviving relay in a mesh does
  while it still holds a route through a relay that went away. The refusal is
  now treated as the peer's tracks ending, so the transport retries and redials
  as it does for any other relay loss.
  (PR [#5892](https://github.com/pipecat-ai/pipecat/pull/5892))

- Fixed `RTVIObserver` remembering the IDs of frames it does not handle, such
  as every audio frame. Its memory still grows with the frames it handles, but
  much more slowly.
  (PR [#5906](https://github.com/pipecat-ai/pipecat/pull/5906))

- Fixed `on_user_turn_idle` never firing after a user turn that ended with no
  transcript, since that turn cancelled the idle timer and nothing restarted
  it.
  (PR [#5907](https://github.com/pipecat-ai/pipecat/pull/5907))

- Fixed observers keeping the IDs of every frame they had seen for the life of
  the session, the pipeline worker's idle detection included.
  (PR [#5908](https://github.com/pipecat-ai/pipecat/pull/5908))

- The `runner` extra now requires `pipecat-ai-prebuilt>=1.2.2`. The
  conversation panel in the prebuilt client UI served by the development runner
  now keeps autoscrolling while long bot replies stream in.
  (PR [#5917](https://github.com/pipecat-ai/pipecat/pull/5917))

### Performance

- `RTVIObserver` now skips audio frames before checking anything else when
  audio levels are not reported, instead of testing every condition on every
  push.
  (PR [#5906](https://github.com/pipecat-ai/pipecat/pull/5906))

## [1.11.0] - 2026-09-17

### Added

- Added `BaseAudioResampler.flush()` and `BaseAudioResampler.reset()`, so
  callers can mark stream boundaries themselves rather than relying on
  `SOXRStreamAudioResampler`'s inactivity timeout. `flush()` returns the audio
  the resampler is still holding, for a stream that has ended; `reset()`
  discards it, for one that was abandoned. Both default to no-ops for
  resamplers that keep no state between calls.
  (PR [#5386](https://github.com/pipecat-ai/pipecat/pull/5386))

- Scripted eval expectations on `function_call` accept an `eval:`: each matched
  call is judged by name and arguments, under a judge prompt of its own, so a
  scenario can check what `args:` cannot match verbatim ("a session about
  OpenTelemetry tracing, submitted for Jennifer Smith"). A rejected call fails
  the turn with kind `judge_no`. `EvalJudge` gains `evaluate_call()`.
  (PR [#5713](https://github.com/pipecat-ai/pipecat/pull/5713))

- `AWSTranscribeSTTService.Settings` gained `partial_results_stability`, which
  selects AWS Transcribe's `"high"`, `"medium"` or `"low"` interim-result
  stability. It still defaults to `"high"`.
  (PR [#5743](https://github.com/pipecat-ai/pipecat/pull/5743))

- Added an `effect` setting to `AzureTTSService` and `AzureHttpTTSService`,
  which passes Azure's `<voice effect>` audio effect processor (`eq_car`,
  `eq_telecomhp8k`) through to the synthesis request.
  (PR [#5744](https://github.com/pipecat-ai/pipecat/pull/5744))

- Added a `reasoning_effort` field to `GroqLLMService.Settings`, which passes
  Groq's reasoning effort control through to the completion request. Accepted
  values vary by model: `"low"`, `"medium"` and `"high"` on the GPT-OSS models
  and `qwen/qwen3.8-27b`, `"none"` and `"default"` on the Qwen models. Setting
  `"none"` also keeps a Qwen model's `<think>` reasoning out of the spoken
  response.
  (PR [#5748](https://github.com/pipecat-ai/pipecat/pull/5748))

- Added a `delay` field to `OpenAIRealtimeSTTService.Settings`, OpenAI's
  latency-versus-accuracy control for how long the model waits before emitting
  transcription text (`minimal`, `low`, `medium`, `high`, `xhigh`). Supported
  by `gpt-realtime-whisper`.
  (PR [#5750](https://github.com/pipecat-ai/pipecat/pull/5750))

- Added a `keywords` field to `OpenAISTTService.Settings`, which passes product
  names, acronyms and other specialized terms to OpenAI's transcription request
  as hints. Supported by `gpt-transcribe`.
  (PR [#5750](https://github.com/pipecat-ai/pipecat/pull/5750))

- Added `math_notation` to `SmallestTTSService.Settings`, Smallest's opt-in
  flag for reading digit-flanked math operators as words (`2 + 2` as "two plus
  two").
  (PR [#5751](https://github.com/pipecat-ai/pipecat/pull/5751))

- Added an `initial_prompt` field to `WhisperSTTService.Settings`, which gives
  Faster Whisper preceding context for each segment, steering the transcript's
  style, punctuation and spelling.
  (PR [#5755](https://github.com/pipecat-ai/pipecat/pull/5755))

- `SarvamRealtimeSTTService` now accepts Sarvam's second-generation realtime
  model, `saaras:v4`, via
  `settings=SarvamRealtimeSTTService.Settings(model="saaras:v4")`.  The default
  remains `saaras:v3-realtime`.
  (PR [#5758](https://github.com/pipecat-ai/pipecat/pull/5758))

- Added video output to `LiveKitTransport`. With `video_out_enabled`, the
  transport publishes a camera track from `RGB`, `RGBA`, `BGRA` or `ARGB`
  frames. `video_out_codec` selects the codec, and the new
  `LiveKitParams.video_out_max_bitrate` caps the bitrate together with
  `video_out_framerate`.
  (PR [#5770](https://github.com/pipecat-ai/pipecat/pull/5770))

- Added a `bot-llm-marker` RTVI message that `RTVIObserver` sends when its new
  `bot_llm_marker_enabled` parameter is on (off by default), reporting the
  turn-completion marker the bot's LLM produced and what it meant: `complete`,
  `short` or `long`.
  (PR [#5777](https://github.com/pipecat-ai/pipecat/pull/5777))

- Scripted eval scenarios can assert on the turn-completion marker the bot's
  LLM produced with a new `llm_marker` event and a `marker:` field naming its
  meaning: `complete`, `short`, `long`, or `incomplete` for either of the last
  two. The bot reports its markers only when a scenario asks for them, so
  clients never see them by default.

    ```yaml
    - user: "Let me think about it, hmmm"
      expect:
        - event: llm_marker
          marker: incomplete
    ```
  (PR [#5777](https://github.com/pipecat-ai/pipecat/pull/5777))

- A scenario file lists its scenarios under `scenarios:`, so one file can hold
  many short cases of the same behavior. Any key a scenario can have may sit at
  the top of the file as the default for all of them, and a scenario that sets
  the same key replaces it whole. `turns:` at the top with one entry per judge
  or modality runs the same conversation under each; `persona:` at the top with
  a `goal:` per entry sends the same caller on different errands. Scripted and
  simulated scenarios can share a file. Each scenario runs on its own, against
  its own bot, and is named `<file>/<scenario>`; the suite's `-s` filter
  accepts that name or either half of it. `EvalScenarioFile.load()` reads a
  file and holds every scenario in it.

    ```yaml
    name: turn_completion
    judge: !include ../judge_text.yaml

    scenarios:
      - name: short_answer
        turns:
          - user: "Japan."
            expect:
              - event: response
      - name: cutoff
        turns:
          - user: "I'd go to Japan because"
            expect:
              - event: response
                absent: true
                within_ms: 3000
    ```
  (PR [#5780](https://github.com/pipecat-ai/pipecat/pull/5780))

- `GeminiLiveLLMService` supports `gemini-3.8-live` and
  `gemini-3.8-live-extended-thinking`, including `NON_BLOCKING` function
  calling (the family's default). Synchronous tools are declared `BLOCKING` on
  `gemini-3.8-live`; `gemini-3.8-live-extended-thinking` accepts only
  `NON_BLOCKING` and logs a warning for them. The thinking model requires a
  `thinking_level` and gets `LOW` when none is configured.
  (PR [#5781](https://github.com/pipecat-ai/pipecat/pull/5781))

- `GeminiLiveLLMService` holds the bot turn open while Gemini reports
  `interaction_status: IN_PROGRESS`, so a reply that spans several
  `turn_complete` messages is recorded as a single assistant turn. Requires
  `google-genai>=2.19.0` (now the `google` extra's floor); older SDKs log a
  warning.
  (PR [#5781](https://github.com/pipecat-ai/pipecat/pull/5781))

- Added `name:` to a manifest entry, its label in the display, the `-p` filter,
  `results.jsonl` (a new `name` field beside `bot`) and the artifact file
  names, which carry the bot path and then the name. It defaults to the `bot:`
  path, for entries that share a bot and differ only in their `runner_body:`.
  Two entries may no longer run the same scenario under one label.
  (PR [#5786](https://github.com/pipecat-ai/pipecat/pull/5786))

- Added `concurrency:` to a manifest entry, capping how many runs of that bot
  execute at once under the suite's own concurrency, for providers that
  rate-limit concurrent connections.
  (PR [#5786](https://github.com/pipecat-ai/pipecat/pull/5786))

- Added `text_excludes:` to a scripted scenario's expectations, the mirror of
  `text_contains`: the event's text must not hold the given substring, for a
  marker the LLM let slip into its reply. It fails with kind `text_present`.
  (PR [#5786](https://github.com/pipecat-ai/pipecat/pull/5786))

- Added `data:` to a manifest entry's `runner_body:`, the bot's runner-args
  body written inline as a mapping; the suite writes it to a file and passes it
  as `--runner-body`. The runner reads a body file as YAML, so it may be YAML
  or JSON.
  (PR [#5786](https://github.com/pipecat-ai/pipecat/pull/5786))

- Added `EvalScriptTurnResult.expectations`, one `EvalExpectationResult` per
  expectation the turn resolved, with whether it passed and what it matched
  (the marker of an `llm_marker`, a function call's signature, a reply's text).
  The suite's `results.jsonl` records carry it too.
  (PR [#5786](https://github.com/pipecat-ai/pipecat/pull/5786))

- Added `LLMMarkerResponseFrame`, pushed by `UserTurnCompletionLLMServiceMixin`
  when a response ends with the raw text the LLM produced, the marker it read
  and the markers it recognizes. The RTVI observer sends it as `bot-llm-marker`
  when `bot_llm_marker_enabled` is on, so the eval harness's `llm_marker` event
  now carries `raw` and `markers`, and a scripted scenario can check the marker
  protocol with `marker_first`, `markers` and `text_after` on an `llm_marker`
  expectation.
  (PR [#5786](https://github.com/pipecat-ai/pipecat/pull/5786))

- `AssemblyAISTTService`'s `language_codes` steering now accepts Urdu,
  Russian, Korean, Catalan, Galician, Romanian, Estonian, Persian, Cantonese,
  Afrikaans, Marathi, Zulu, Xhosa, and Norwegian Nynorsk, matching the full
  set `universal-3-5-pro` and `universal-3-6-pro` support.
  (PR [#5808](https://github.com/pipecat-ai/pipecat/pull/5808))

- Added `speed` to the eval harness's `speech:` block, Kokoro's rate
  multiplier, so a scenario can make the synthesized user talk faster or
  slower. Cached user audio is keyed by the speed too.
  (PR [#5813](https://github.com/pipecat-ai/pipecat/pull/5813))

### Changed

- `BasetenLLMService` defaults to `zai-org/GLM-5.3`. Set
  `settings=BasetenLLMService.Settings(model=...)` to use a different model.
  (PR [#5730](https://github.com/pipecat-ai/pipecat/pull/5730))

- `DeepgramFluxSTTService` now documents its default `url` as Deepgram's Flux
  endpoint, `wss://api.deepgram.com/v2/listen`.
  (PR [#5739](https://github.com/pipecat-ai/pipecat/pull/5739))

- `DeepSeekLLMService` defaults to `deepseek-flash`, DeepSeek's current name for
  V4.1 Flash. The previous default, `deepseek-v4-flash`, names a retired model
  and is only temporarily routed to V4.1 Flash. Set
  `settings=DeepSeekLLMService.Settings(model=...)` to use a different model.
  (PR [#5746](https://github.com/pipecat-ai/pipecat/pull/5746))

- Added a `mode` field to the OpenAI Responses `ReasoningConfig`, selecting the
  `standard` or `pro` reasoning mode on models that offer one, such as gpt-5.6.
  (PR [#5749](https://github.com/pipecat-ai/pipecat/pull/5749))

- `SonioxTTSService` settings now include `client_reference_id`, the identifier
  Soniox records with each request in its usage logs, matching the field
  `SonioxSTTService` already exposes.
  (PR [#5752](https://github.com/pipecat-ai/pipecat/pull/5752))

- Added `max_session_duration` to `LiveAvatarNewSessionRequest`, capping a
  HeyGen LiveAvatar session at a given number of seconds.
  (PR [#5753](https://github.com/pipecat-ai/pipecat/pull/5753))

- `InworldRealtimeLLMService` docstrings now use `openai/gpt-4.1-mini` in their
  examples instead of `openai/gpt-4.1-nano`, which OpenAI shuts down on
  2026-10-23.
  (PR [#5754](https://github.com/pipecat-ai/pipecat/pull/5754))

- `RimeHttpTTSService` now sends `noTextNormalization` to Rime, so the setting
  takes effect on Mist requests.
  (PR [#5756](https://github.com/pipecat-ai/pipecat/pull/5756))

- `SarvamLLMService` accepts `deepseekv4-flash`, the DeepSeek V4 Flash model
  Sarvam serves on `/v2` with a 1M-token context window, tool calling, and
  reasoning.
  (PR [#5757](https://github.com/pipecat-ai/pipecat/pull/5757))

- `pipecat init` React clients (Vite and Next.js) now use [Pipecat
  UI](https://ui.pipecat.ai) instead of `@pipecat-ai/voice-ui-kit`. The
  generated client renders the Pipecat UI console: connect flow, transcript,
  metrics, device and session info, and a live event stream. Components are
  installed as source under `src/components/pipecat`, and `components.json`
  points at the `@pipecat` registry, so `npx shadcn@latest add
  @pipecat/<component>` adds more. The vanilla JavaScript client is restyled to
  match.
  (PR [#5767](https://github.com/pipecat-ai/pipecat/pull/5767))

- `pipecat eval suite` now takes its queue round-robin across the manifest's
  entries within each attempt, so a slow or rate-limited provider holds only
  its share of the concurrency instead of every slot until its scenarios are
  done.
  (PR [#5768](https://github.com/pipecat-ai/pipecat/pull/5768))

- Rewrote the default instructions that `filter_incomplete_user_turns` appends
  to the system prompt, so weaker models follow the ●/◐/○ marker protocol more
  reliably. Short answers such as "yes" or "Tuesday" are now called out as
  complete, a mid-sentence fragment is explicitly not a request for help, a
  continuation is judged together with the fragment before it, and the examples
  are prose rather than arrows or a transcript, since small models reproduce
  whatever shape they are shown. Bots that set
  `UserTurnCompletionConfig.instructions` are unaffected.
  (PR [#5768](https://github.com/pipecat-ai/pipecat/pull/5768))

- `SpeechmaticsSTTService` now defaults `turn_detection_mode` to
  `TurnDetectionMode.EXTERNAL`, so Pipecat's own VAD drives turn boundaries
  through `finalize()`. Set
  `settings=SpeechmaticsSTTService.Settings(turn_detection_mode=TurnDetectionMode.VAD)`
  to keep letting the Speechmatics service run its own VAD and close turns
  itself.
  (PR [#5773](https://github.com/pipecat-ai/pipecat/pull/5773))

- The `runner` extra now requires `pipecat-ai-prebuilt>=1.1.1`. The prebuilt
  client UI served by the development runner moves to `@pipecat-ai/client-js`
  1.13.1, `@pipecat-ai/daily-transport` 1.6.9,
  `@pipecat-ai/small-webrtc-transport` 1.10.8, and
  `@pipecat-ai/websocket-transport` 1.7.2.
  (PR [#5776](https://github.com/pipecat-ai/pipecat/pull/5776))

- The `runner` extra now requires `pipecat-ai-prebuilt>=1.2.1`. The prebuilt
  client UI served by the development runner is built with [Pipecat
  UI](https://ui.pipecat.ai) instead of `@pipecat-ai/voice-ui-kit`, shows the
  bot video pane in the console, and follows the system light/dark theme on
  load.
  (PR [#5799](https://github.com/pipecat-ai/pipecat/pull/5799))

- The `sagemaker` and `aws-nova-sonic` extras now require
  `aws_sdk_sagemaker_runtime_http2>=0.6.0,<0.10` and
  `aws_sdk_bedrock_runtime>=0.6.0,<0.10`. The two SDKs share `smithy-core`, so
  they move together.
  (PR [#5807](https://github.com/pipecat-ai/pipecat/pull/5807))

- Async function calls (`cancel_on_interruption=False`) now keep settling as
  ordinary tool results when only assistant output lands in the context while
  they run, such as filler a tool handler speaks with `TTSSpeakFrame`.
  Previously that filler turned the result into a deferred-result message,
  while the same filler spoken from `on_function_calls_started` did not. The
  result is still deferred when a user or developer message arrives first, or
  when the call sent an intermediate update.
  (PR [#5820](https://github.com/pipecat-ai/pipecat/pull/5820))

- `resolve_language(..., use_base_code=True)` now resolves a regional variant
  missing from the map to the map's code for its base language when the map has
  one (`en-US` becomes `"eng"` for a map with `Language.EN: "eng"`), falling
  back to the base code (`"en"`) otherwise. This changes the code sent for
  unmapped regional variants by Meta STT, Pocket TTS, Rime TTS, Camb TTS,
  Kokoro TTS, XTTS, Neuphonic TTS, and xAI TTS; for example, Rime sends `"ara"`
  for `Language.AR_AE` instead of `"ar"`.
  (PR [#5824](https://github.com/pipecat-ai/pipecat/pull/5824))

- `pipecat eval suite` runs each manifest entry's scenarios back to back on one
  concurrency slot, so an entry's next scenario starts as soon as its previous
  one ends and a slow provider holds no more than one slot. A repeated sweep is
  attempt-major, every entry's first attempt before any entry's second. An
  entry's `concurrency:` now means how many slots it may hold at once, one by
  default.
  (PR [#5825](https://github.com/pipecat-ai/pipecat/pull/5825))

### Deprecated

- Deprecated `EvalSimulationDriver.transcript()`. No replacement: the driver
  feeds the judge as the conversation happens. Will be removed in 2.0.0.
  (PR [#5765](https://github.com/pipecat-ai/pipecat/pull/5765))

- Deprecated the `transcript` parameter of `EvalJudge.evaluate_run()`. The
  judge keeps the conversation it judges: feed it with `add_user_message`,
  `add_assistant_message` and the new `add_tool_call`, and call
  `evaluate_run(criteria, success)`. Will be removed in 2.0.0.
  (PR [#5765](https://github.com/pipecat-ai/pipecat/pull/5765))

- `EvalScriptScenario.load()` and `EvalSimulationScenario.load()` are
  deprecated and will be removed in 2.0.0. Use `EvalScenarioFile.load()`, which
  reads a scenario file of either kind.
  (PR [#5780](https://github.com/pipecat-ai/pipecat/pull/5780))

- `load_scenario_file()` is deprecated and will be removed in 2.0.0. Use
  `EvalScenarioFile.load()`, which reads a file and holds every scenario in it.
  (PR [#5780](https://github.com/pipecat-ai/pipecat/pull/5780))

- A scenario file whose top level holds `turns:` or `persona:` instead of a
  `scenarios:` list is deprecated and will stop loading in 2.0.0. It still
  loads as one scenario under the file's name, with a `DeprecationWarning`.
  Wrap the scenario in a `scenarios:` list with its own `name:`.
  (PR [#5780](https://github.com/pipecat-ai/pipecat/pull/5780))

- Deprecated a manifest entry's bare `runner_body: <file>`. Use `runner_body:
  {path: <file>}` instead. Will be removed in 2.0.0.
  (PR [#5786](https://github.com/pipecat-ai/pipecat/pull/5786))

### Fixed

- Fixed the output transport dropping speech whenever TTS pauses for more than
  200 ms between chunks. This only affected pipelines where the TTS render rate
  differs from the transport's output rate.
  (PR [#5386](https://github.com/pipecat-ai/pipecat/pull/5386))

- Fixed a word-timestamp event that belongs to no sentence being discarded with
  nothing logged, in streaming mode (a TTS service using
  `TextAggregationMode.TOKEN`). This comes up when a provider reports a
  punctuation mark as its own event after the preceding word already took it.
  Such an event used to sit in a buffer until the next turn began and was
  thrown away there; it is now logged and dropped when its own turn ends.
  (PR [#5681](https://github.com/pipecat-ai/pipecat/pull/5681))

- Fixed `AggregatedFrameSequencer.force_complete` emitting the text it
  force-completes as a `TTSTextFrame` without a matching
  `AggregatedTextProgressFrame`, so the progress view stopped where the TTS
  provider stopped reporting words while the word frames carried the rest of
  the turn.
  (PR [#5681](https://github.com/pipecat-ai/pipecat/pull/5681))

- Fixed audio eval sessions remaining silent after a text-mode session on the
  same bot by resetting `skip_tts` for each connection.
  (PR [#5722](https://github.com/pipecat-ai/pipecat/pull/5722))

- A failed `MCPClient` tool call now gives the LLM the error the call raised,
  cut to 200 characters. The fixed line `Sorry, could not call the mcp tool`
  remains only for a call that raised nothing and returned no text content.
  Code that matched on the old text must read the new one.
  (PR [#5734](https://github.com/pipecat-ai/pipecat/pull/5734))

- `CartesiaSTTService` now sends `keyterm` for Cartesia's `ink-preview` models,
  which support keyterms alongside `ink-2`.
  (PR [#5738](https://github.com/pipecat-ai/pipecat/pull/5738))

- Corrected the documented range of the Deepgram Flux TTS `speed` setting: Flux
  accepts 0.5 to 1.5 in steps of 0.05, not 0.85 to 1.15.
  (PR [#5740](https://github.com/pipecat-ai/pipecat/pull/5740))

- Fixed `GoogleLLMService` failing every request with `gemini-3.8-flash`. That
  model rejects the `minimal` thinking level Pipecat applies by default to
  Gemini 3 Flash models, so it now gets `low`, the lowest level it accepts.
  (PR [#5741](https://github.com/pipecat-ai/pipecat/pull/5741))

- The Mem0 RAG example's local-configuration snippet now names
  `claude-sonnet-4-6` instead of the retired `claude-3-5-sonnet-20240620`.
  (PR [#5742](https://github.com/pipecat-ai/pipecat/pull/5742))

- Corrected `MoonshineSTTService`'s licensing note: Moonshine's models are MIT
  in every language and size, except the legacy non-streaming (`tiny`, `base`)
  non-English models, which stay under the non-commercial Moonshine Community
  License.
  (PR [#5745](https://github.com/pipecat-ai/pipecat/pull/5745))

- Updated the Fireworks update-settings example to use
  `nemotron-3-ultra-nvfp4`; Fireworks retired `gpt-oss-20b` from serverless.
  (PR [#5747](https://github.com/pipecat-ai/pipecat/pull/5747))

- The OpenAI Responses services no longer send `reasoning.effort="none"` for
  `gpt-6-astra`, which rejects that value. The model now works out of the box.
  (PR [#5749](https://github.com/pipecat-ai/pipecat/pull/5749))

- Fixed `HeyGenVideoService` avatar state logging, which watched for the
  LiveAvatar LITE event `agent.state`. LiveAvatar renamed it to
  `agent.state_updated`.
  (PR [#5753](https://github.com/pipecat-ai/pipecat/pull/5753))

- Fixed the `voice-sarvam` example's optional voice-switch line, which named
  `anushka`, a `bulbul:v2` speaker that `bulbul:v3` rejects. It now names
  `anand`, a `bulbul:v3` speaker.
  (PR [#5759](https://github.com/pipecat-ai/pipecat/pull/5759))

- Fixed per-tool `timeout_secs` (and global `function_call_timeout_secs`) being
  disarmed by an intermediate `FunctionCallResultProperties(is_final=False)`
  update. An async tool that reported progress no longer cancels its own
  deadline; only a final result clears it.
  (PR [#5760](https://github.com/pipecat-ai/pipecat/pull/5760))

- `npm run lint` in a generated React client failed before linting any file,
  because the ESLint config used a legacy plugin shape that ESLint 10 rejects.
  The Vite template now uses the flat react-hooks config, and the Next.js
  template pins ESLint 9 since `eslint-config-next` 16 does not run on 10.
  (PR [#5767](https://github.com/pipecat-ai/pipecat/pull/5767))

- Fixed `SambaNovaLLMService` ignoring `filter_incomplete_user_turns`. Its text
  bypassed the turn-completion mixin, so incomplete turns were never held and
  the marker character was spoken by TTS.
  (PR [#5768](https://github.com/pipecat-ai/pipecat/pull/5768))

- Fixed `LiveKitTransport` reporting a disconnect twice when it disconnected,
  and reporting one for a connection attempt that failed.
  (PR [#5770](https://github.com/pipecat-ai/pipecat/pull/5770))

- Fixed `LiveKitTransport` leaving its audio track unpublished when a track
  publish failed while connecting and the connection was retried.
  (PR [#5770](https://github.com/pipecat-ai/pipecat/pull/5770))

- Fixed `LiveKitTransport` not releasing its outgoing audio source when it
  disconnected.
  (PR [#5770](https://github.com/pipecat-ai/pipecat/pull/5770))

- Fixed `GladiaSTTService` tearing down and reconnecting its websocket when
  Gladia reports a translation addon failure, and silently ignoring rejected
  audio chunks. Both are now reported upstream as non-fatal `ErrorFrame`s while
  transcription continues on the same connection.
  (PR [#5772](https://github.com/pipecat-ai/pipecat/pull/5772))

- `MCPClient` now drops an MCP session that a tool call found dead. The next
  tool call connects again. A client that was never started, or that its owner
  closed, still raises on a tool call.
  (PR [#5778](https://github.com/pipecat-ai/pipecat/pull/5778))

- Fixed `ElevenLabsDialogueTTSService` closing its WebSocket with a `1008`
  policy violation (`voices can only be set in the first message`), which lost
  the rest of the turn. A context that had gone quiet was dropped locally while
  ElevenLabs was still generating for it, so the next sentence of the turn
  registered that context a second time.
  (PR [#5795](https://github.com/pipecat-ai/pipecat/pull/5795))

- Fixed `VADProcessor` ignoring its `speech_activity_period` while the user is
  speaking.
  (PR [#5805](https://github.com/pipecat-ai/pipecat/pull/5805))

- Fixed SageMaker connections with a connection setting the endpoint rejects
  (such as `language_hint` on `flux-general-en`) hanging until the Flux
  connection timeout. `DeepgramFluxSageMakerSTTService` now reports SageMaker's
  424 error through `on_connection_error` within a second. When the session
  does time out, the error names the endpoint's CloudWatch log group, which
  holds the container's error text.
  (PR [#5807](https://github.com/pipecat-ai/pipecat/pull/5807))

- `AssemblyAISTTService` now raises a clear error as soon as `prompt` is set
  on a non-U3-Pro model, instead of forwarding it and letting the connection
  fail server-side.
  (PR [#5808](https://github.com/pipecat-ai/pipecat/pull/5808))

- Fixed Gemini 3 rejecting a request with `Function call is missing a
  thought_signature` when the context holds function calls Gemini did not
  produce, such as those made by another LLM before switching to Gemini or ones
  added by the application. `GeminiLLMAdapter` now sends Google's documented
  placeholder signature for those calls.
  (PR [#5809](https://github.com/pipecat-ai/pipecat/pull/5809))

- Fixed a realtime service configured with its own tools (such as
  `UltravoxRealtimeLLMService` with `one_shot_selected_tools`) answering a tool
  call with the missing-function result when the model called the tool before
  the first context frame reached the service. The handlers of a service's own
  tools are now registered when the service starts.
  (PR [#5812](https://github.com/pipecat-ai/pipecat/pull/5812))

- Fixed `UltravoxRealtimeLLMService` never reporting an async tool's result
  when the result arrived while the bot was still speaking. The result was sent
  as a second `client_tool_result`, which Ultravox ignores after the
  placeholder it already received; it is now delivered as user-side text like a
  deferred result.
  (PR [#5812](https://github.com/pipecat-ai/pipecat/pull/5812))

- Fixed `SarvamSTTService` transcripts arriving with `finalized=False`, which
  made the turn analyzer wait out its timeout for more transcript before ending
  the user turn. Each transcript is now marked finalized, so the turn can end
  as soon as the transcript arrives.
  (PR [#5815](https://github.com/pipecat-ai/pipecat/pull/5815))

- Fixed `GoogleLLMService` (and `GoogleVertexLLMService`) failing with a 400
  (`Requests ending with a model turn are not supported.`) on newer Gemini
  models, such as the default `gemini-3.6-flash`, when the context ends with an
  assistant message. This happens when a tool handler speaks filler with
  `TTSSpeakFrame` before its result arrives. A minimal `"."` user turn is now
  appended to the request in that case, for every model not known to continue a
  trailing model turn; the stored context is unchanged.
  (PR [#5821](https://github.com/pipecat-ai/pipecat/pull/5821))

- `MiniMaxHttpTTSService` now resolves a regional language variant to its base
  language's name for `language_boost` (`Language.PT_BR` becomes
  `"Portuguese"`) instead of sending the raw `pt-BR` code.
  (PR [#5824](https://github.com/pipecat-ai/pipecat/pull/5824))

- `ElevenLabsRealtimeSTTService` now converts its `language` setting to
  ElevenLabs' language codes, so a regional variant such as `Language.EN_US`
  connects with `language_code=eng` instead of being rejected with a 1008
  `invalid_request` close. `ElevenLabsSTTService` also resolves regional
  variants to their base language's code rather than sending `en-US`.
  (PR [#5824](https://github.com/pipecat-ai/pipecat/pull/5824))

- Fixed the eval harness scoring bot speech from before the user's turn as the
  reply, or losing the reply behind it, when that speech was transcribed late.
  A send now waits until everything the bot said is transcribed, and a turn
  that talks over the bot drops the transcripts of what it was saying, by when
  their audio began.
  (PR [#5825](https://github.com/pipecat-ai/pipecat/pull/5825))

- Fixed an `absent: true` response expectation failing on a later sentence of
  the reply an earlier expectation had already matched. It now counts only a
  reply the bot began after that match.
  (PR [#5825](https://github.com/pipecat-ai/pipecat/pull/5825))

### Performance

- Reduced per-frame memory work for rolling audio volume tracking by retaining
  the window in a mutable buffer.
  (PR [#5811](https://github.com/pipecat-ai/pipecat/pull/5811))

## [1.10.0] - 2026-09-11

### Added

- `SmallestTTSService` now uses Smallest AI's continuation API: text fragments
  within the same LLM turn share a `context_id` so the server joins them into
  one continuous generation instead of resetting prosody on each request. Added
  an optional `max_buffer_delay_ms` setting to control the server-side
  buffering window.
  (PR [#5626](https://github.com/pipecat-ai/pipecat/pull/5626))

- Added `LiveKitParams.audio_out_queue_size_ms` to configure the outgoing
  `rtc.AudioSource` buffer size. Defaults to LiveKit's 1000 ms, so existing
  behaviour is unchanged.
  (PR [#5699](https://github.com/pipecat-ai/pipecat/pull/5699))

- Added `enable_turn_detection` to `GradiumSTTService`. With it on, Gradium's
  server-side end-pointing signal decides when user turns start and end instead
  of the pipeline's VAD, and the service recommends
  `ExternalUserTurnStrategies` to the user aggregator. The new `eot_horizon_s`,
  `eot_threshold` and `post_flush_cooldown_frames` settings tune it.
  (PR [#5706](https://github.com/pipecat-ai/pipecat/pull/5706))

### Changed

- ⚠️ **Behavior change:** Pipecat now supports the openai 3 SDK, and the
  `openai` dependency is widened to `>=1.74.0,<4`, so a fresh resolve picks
  openai 3. It builds its HTTP clients on `httpx2` instead of `httpx`, and TLS
  certificates then verify against the operating system trust store rather than
  `certifi`. Minimal container images without system CA certificates, and
  environments behind a TLS-inspecting proxy, may need `SSL_CERT_FILE` or
  `SSL_CERT_DIR` pointed at a CA bundle. Pin `openai<3` to stay on the previous
  HTTP stack.

      A `Timeout` passed to a service's `http_client` — `OpenAITTSService` and
  the Whisper-based STT services — must come from the HTTP client family the
  installed SDK uses: `httpx2` on openai 3, `httpx` before it.
  (PR [#5620](https://github.com/pipecat-ai/pipecat/pull/5620))

- Widened the `anthropic` dependency to `>=0.49.0,<2` to support the anthropic
  1 SDK. `AnthropicLLMService` sends `temperature`, `top_k` and `top_p` through
  the request's `extra_body`, since the Messages API methods dropped them as
  parameters in anthropic 1. Requests reach the API unchanged, and the settings
  keep their names and meaning.

      If you pass your own Bedrock client, anthropic 1 requires an explicit
  region — `AsyncAnthropicBedrock(aws_region=...)`, or `AWS_REGION` in the
  environment — where it previously fell back to `us-east-1`.
  (PR [#5623](https://github.com/pipecat-ai/pipecat/pull/5623))

- Widened the `mcp` dependency to `mcp[cli]>=1.24.0,<3` so `MCPClient` works
  with the MCP SDK's 2.x line as well as 1.x. The floor moves to 1.24.0, the
  first release carrying the `streamable_http_client` transport that both lines
  share.
  (PR [#5624](https://github.com/pipecat-ai/pipecat/pull/5624))

- `SpeechmaticsSTTService` reconnects with exponential backoff after a dropped
  connection or a recoverable server error, buffering audio meanwhile. A
  rejected credential, rejected session, request-rejecting server error, or
  exhausted reconnect attempts are reported as a permanent error, leaving the
  service unusable for the `PipelineWorker`'s `ProcessorUnusablePolicy` to act
  on. A settings update that requires a reconnect gives a rejected session
  another chance.
  (PR [#5631](https://github.com/pipecat-ai/pipecat/pull/5631))

- ⚠️ **Breaking change:** `SpeechmaticsSTTService` now targets Speechmatics
  Agent STT (`/v2/agent`) through the `speechmatics-agent-stt` SDK, which
  replaces `speechmatics-voice[smart]` in the `speechmatics` extra. The service
  cannot connect to the legacy real-time endpoint. The default
  `turn_detection_mode` is now `TurnDetectionMode.VAD`, so Speechmatics closes
  turns server-side and the service recommends `ExternalUserTurnStrategies`;
  pass `turn_detection_mode=TurnDetectionMode.EXTERNAL` to keep driving turns
  from Pipecat's own VAD. The default model is `linden-1`, selected by the new
  `model` setting. `Settings.include_partials` is renamed `enable_partials`.
  (PR [#5631](https://github.com/pipecat-ai/pipecat/pull/5631))

- The `pipecat eval suite` live dashboard now shows a repeated (bot, scenario)
  row's pass rate while its attempts are still running, beside how many are
  left, instead of only once every attempt is in. A pass rate, on the dashboard
  and in the piped summary, is green while every finished attempt passed and
  red once one has failed; the yellow middle band is gone.
  (PR [#5704](https://github.com/pipecat-ai/pipecat/pull/5704))

- The `recording` line in `pipecat eval suite`'s settings now says which of the
  selected runs will actually record. Only an audio-mode run produces audio, so
  with recording on it reads `on (3 of 6 runs; text mode skipped)` when the
  selection is mixed and `off (all runs text mode)` when nothing will be
  recorded. `pipecat eval run -a` prints the same line.
  (PR [#5707](https://github.com/pipecat-ai/pipecat/pull/5707))

- `DeepSeekLLMService` now defaults `thinking` to disabled. DeepSeek's V4
  models otherwise reason before every answer, which delays the first spoken
  token. Pass `thinking=DeepSeekLLMService.ThinkingConfig(type="enabled")` to
  turn it on, or `thinking=None` to use DeepSeek's own default.
  (PR [#5710](https://github.com/pipecat-ai/pipecat/pull/5710))

### Deprecated

- Deprecated `SpeechmaticsSTTService` `operating_point` on `Settings` and
  `InputParams`; use `model` instead. Passing it still selects the model and
  emits a `DeprecationWarning`. It will be removed in 2.0.0.
  (PR [#5631](https://github.com/pipecat-ai/pipecat/pull/5631))

### Removed

- ⚠️ **Breaking change:** Agent STT has no speaker focus, so
  `SpeechmaticsSTTService` no longer exports `SpeakerFocusMode` or
  `SpeakerFocusConfig`, and `UpdateParams` and `update_params()` are removed.
  The `Settings` and `InputParams` fields `focus_speakers`, `ignore_speakers`,
  `focus_mode`, `speaker_passive_format`, `max_delay`,
  `end_of_utterance_silence_trigger`, `end_of_utterance_max_delay`,
  `split_sentences`, `include_results`, and `extra_params` are removed, as are
  `TurnDetectionMode.FIXED`, `ADAPTIVE`, and `SMART_TURN`. `OperatingPoint` is
  replaced by `Model`.
  (PR [#5631](https://github.com/pipecat-ai/pipecat/pull/5631))

- ⚠️ Removed `LmntTTSService`. LMNT has shut down and is no longer available.
  (PR [#5698](https://github.com/pipecat-ai/pipecat/pull/5698))

### Fixed

- Fixed the WebSocket transports treating an audio frame as unsent whenever the
  serializer emitted no payload for it. A serializer that resamples through a
  stream resampler buffers audio across calls and emits it in a later one, so
  on a pipeline whose output sample rate differs from the wire rate most frames
  took that path. Those frames are now paced and pushed downstream like any
  other, so the output queue drains at playback speed again and
  `BotStoppedSpeakingFrame` and the `EndFrame` behind it are no longer reached
  early, which matters most where the serializer hangs up the call on the
  `EndFrame`.
  (PR [#5593](https://github.com/pipecat-ai/pipecat/pull/5593))

  Fixed `PIPECAT_SETUP_FILES` parsing on platforms whose path separator is not a
  colon.
  (PR [#5638](https://github.com/pipecat-ai/pipecat/pull/5638))

- Fixed the `pipecat eval suite` live dashboard showing a repeated (bot,
  scenario) row as idle while a finished attempt's bot was still being stopped.
  The row now keeps spinning until its concurrency slot is released. `EvalRun`
  gains a `stopping` flag for that window.
  (PR [#5704](https://github.com/pipecat-ai/pipecat/pull/5704))

- Async function calls (`cancel_on_interruption=False`) whose result arrives
  before the conversation moves on are now recorded as ordinary tool results,
  without the deferred-result message that asks the LLM to convey them. The
  deferred delivery still applies when the LLM has responded, a new message has
  landed, or the call sent an intermediate update in the meantime.
  (PR [#5705](https://github.com/pipecat-ai/pipecat/pull/5705))

- Fixed `DeepSeekLLMService` failing every request after a tool call in
  thinking mode with a 400 (`The reasoning_content in the thinking mode must be
  passed back to the API`). DeepSeek requires `reasoning_content` on each
  assistant message of the current turn once a tool call is involved; the new
  `DeepSeekLLMAdapter` supplies an empty one on assistant messages that have
  none.
  (PR [#5710](https://github.com/pipecat-ai/pipecat/pull/5710))

## [1.9.0] - 2026-09-10

### Added

- Added `LatencyBreakdown.contributions`, a timeline of the user-to-bot
  interval whose durations sum to the measured latency. It names the time no
  service reports — VAD silence, turn detection, turn-completion markers and
  holds, sentence aggregation, and function handlers — and attributes time a
  setting governs to that setting rather than to the service it runs inside.

  The first thing the bot says is measured too, from where
  `StartupTimingObserver` stops — the pipeline ready with a client on it — so
  the two reports meet rather than overlap. Waiting to be asked is usually most
  of that wait, and no service reports it.

  Each part carries a stable `key` and an `owner_kind` — service, setting, bot
  or pipeline — so data keyed on them survives a label being reworded, and the
  breakdown says what it was `measured_from` alongside the `total_secs` it
  measured. Print it with `LatencyBreakdown.turn_contribution_lines()`;
  `UserBotLatencyObserver(min_contribution_secs=...)` sets how brief a part
  must be before it rolls into a single `pipeline` entry.

      ```
      0.200s  endpointing wait     [config: VAD stop_secs]
      0.125s  transcription        [DeepgramSTTService#0]
      0.336s  LLM inference        [OpenAILLMService#0]
      0.024s  turn completion      [config: filter_incomplete_user_turns]
      0.359s  speech synthesis     [CartesiaTTSService#0]
      1.044s  TOTAL
      ```
  (PR [#5445](https://github.com/pipecat-ai/pipecat/pull/5445))

- Eval scenario turns can play an audio file as the user instead of
  synthesizing text. A turn's `audio:` names a recording, resolved relative to
  the scenario file, in any format `soundfile` reads (WAV, MP3, FLAC, OGG,
  ...); multi-channel audio is downmixed to mono and the file keeps its own
  sample rate. `user:` is required alongside it and is what the recording says,
  so the judge and `text_contains` still have the turn's input. A scenario
  whose audio turns all name a file needs no `user.speech:` block.
  (PR [#5473](https://github.com/pipecat-ai/pipecat/pull/5473))

- Added a `cancellable_by_llm` argument to `LLMSwitcher.register_function()`,
  which forwards it to every LLM the switcher fronts. Tools registered through
  a switcher can now opt into LLM-side cancellation, matching what
  `LLMService.register_function()` accepts.
  (PR [#5474](https://github.com/pipecat-ai/pipecat/pull/5474))

- Added `AssemblyAISyncSTTService`, a segmented speech-to-text service backed
  by AssemblyAI's Sync API: pipeline VAD segments the audio and each segment
  (up to 120 seconds) is transcribed in one HTTP request, with no streaming
  session to manage. It takes an `aiohttp_session`, `base_url` selects a
  data-residency endpoint, and `Settings` supports `model`, `language`,
  `prompt`, `keyterms_prompt`, and `conversation_context`.

  Recent user and agent turns are sent automatically as `conversation_context`
  on each request, bounded by `max_context_turns` (`0` disables) and
  `max_context_chars`; setting `conversation_context` yourself replaces that
  buffer. The connection is pre-warmed when the user starts speaking so the
  transcription request skips the handshake (`enable_prewarming`).
  (PR [#5488](https://github.com/pipecat-ai/pipecat/pull/5488))

- `AzureSTTService.Settings` gained `segmentation_silence_timeout_ms`, which
  sets how much silence (100–5000 ms) Azure allows inside a phrase before it
  emits a final transcript. Azure's default of 500 ms applies when it's unset.
  (PR [#5511](https://github.com/pipecat-ai/pipecat/pull/5511))

- Added `on_progress` and `on_update` event handlers to the eval framework.
  `EvalSession` and `EvalSuite` are now `BaseObject`s, so eval progress is
  observed the same way as every other Pipecat event. `EvalSuite`'s `on_update`
  handlers run synchronously and should return promptly, since an `EvalRun` is
  mutated in place as it executes; `EvalSession`'s `on_progress` handlers run
  as tasks, and `EvalSession.run()` waits for them before it returns.

        @session.event_handler("on_progress")
        async def on_progress(session, progress):
            print(progress.event_name, progress.status)
  (PR [#5520](https://github.com/pipecat-ai/pipecat/pull/5520))

- Added a `voice_parameters` setting to `AzureTTSService` and
  `AzureHttpTTSService`, which passes SSML's `parameters` attribute on
  `<voice>` through to Azure so HD voices can be tuned (`temperature`, `top_p`,
  `top_k`, `cfg_scale`, `enhancePronunciation`).
  (PR [#5522](https://github.com/pipecat-ai/pipecat/pull/5522))

- Added `profanity_filter` and `redact` to `DeepgramFluxSTTService.Settings`
  and `DeepgramFluxSageMakerSTTService.Settings`, exposing Flux's profanity
  masking and number redaction.
  (PR [#5523](https://github.com/pipecat-ai/pipecat/pull/5523))

- Added a `version` field to `DeepgramSTTService.Settings` (and
  `DeepgramSageMakerSTTService.Settings`), pinning transcription to a specific
  Deepgram model version instead of whichever one `latest` currently resolves
  to. Previously reachable only as an untyped `extra` key, which continues to
  work.
  (PR [#5527](https://github.com/pipecat-ai/pipecat/pull/5527))

- Added `thinking` to `DeepSeekLLMService.Settings`, which controls DeepSeek's
  thinking mode: `settings=DeepSeekLLMService.Settings(thinking={"type":
  "disabled"})` turns off the reasoning pass that V4 models run by default.
  (PR [#5528](https://github.com/pipecat-ai/pipecat/pull/5528))

- `CartesiaTurnsSTTSettings` gained `turn_start_threshold`,
  `turn_eager_end_threshold`, `turn_end_threshold` and `turn_end_timeout_ms`,
  Cartesia's turn detection tuning parameters. Unset by default, so the
  server's own defaults apply.
  (PR [#5533](https://github.com/pipecat-ai/pipecat/pull/5533))

- Added a `denoiser_config` setting to `GoogleSTTService`, which passes
  Google's Chirp 3 background-noise removal (`denoise_audio`, `snr_threshold`)
  through to the recognition request.
  (PR [#5536](https://github.com/pipecat-ai/pipecat/pull/5536))

- Added a `temperature` field to `HumeTTSService.Settings`, passing Hume's
  sampling temperature through to the synthesis request. Higher values increase
  variation, lower values increase consistency; when unset, Hume applies its
  own per-model default.
  (PR [#5540](https://github.com/pipecat-ai/pipecat/pull/5540))

- Added a `speed` setting to `DeepgramTTSService` and `DeepgramHttpTTSService`,
  passing Deepgram's Aura speech-rate multiplier (0.7 to 1.5) through to
  `/v1/speak`.
  (PR [#5552](https://github.com/pipecat-ai/pipecat/pull/5552))

- Added a `no_verbatim` setting to `ElevenLabsSTTService` and
  `ElevenLabsRealtimeSTTService`, which asks Scribe to drop filler words, false
  starts and non-speech sounds from the transcript.
  (PR [#5555](https://github.com/pipecat-ai/pipecat/pull/5555))

- Added a `hotwords` field to `WhisperSTTService.Settings`, which biases Faster
  Whisper's transcription towards the words or phrases it names, such as
  product names or domain jargon.
  (PR [#5558](https://github.com/pipecat-ai/pipecat/pull/5558))

- Added a `reduce_silence` setting to `SonioxTTSService`, which shortens the
  pauses between words on models that support silence reduction.
  (PR [#5567](https://github.com/pipecat-ai/pipecat/pull/5567))

- Added an `include_results` setting to `SpeechmaticsSTTService`, which asks
  Speechmatics to include word-level results in transcript messages.
  (PR [#5568](https://github.com/pipecat-ai/pipecat/pull/5568))

- `WebsocketService` now accepts `reconnect_backoff_min_wait` and
  `reconnect_backoff_max_wait` to configure the wait between reconnection
  attempts. Defaults (4s / 10s) preserve the previous hardcoded behavior.
  (PR [#5586](https://github.com/pipecat-ai/pipecat/pull/5586))

- Added `MetaSTTService`, a streaming speech-to-text service using Meta's
  `muse-voice-transcribe-1.0` model over the Muse Voice realtime API. Install
  it with `pipecat-ai[meta]`, or pick Meta when scaffolding with `pipecat
  init`. `MetaSTTService.Settings` takes `language` or `language_bias` to bias
  recognition across the 25 supported languages, `keywords` for domain-specific
  terms, and `mode` to select the model's own endpointing, client-delimited
  turns, or diarization.
  (PR [#5587](https://github.com/pipecat-ai/pipecat/pull/5587))

- Added `universal-3-6-pro` as a supported `AssemblyAISTTService` model. It is
  `universal-3-5-pro` upgraded — the same model with the same feature set — and
  is recognized as part of the Universal-3 Pro family, so every `u3-rt-pro`
  feature (built-in turn detection, prompting, continuous partials,
  `interruption_delay`, context carryover, and voice focus) applies to it as
  well. `universal-3-5-pro` remains supported and stays the default.
  (PR [#5598](https://github.com/pipecat-ai/pipecat/pull/5598))

- `StartupTimingReport.warmup` reports what warming Pipecat's deferred imports
  cost startup, which no processor accounts for and which is often the largest
  single contributor to a cascaded bot's start. `warmup.duration_secs` is how
  long warming took and `warmup.blocking_duration_secs` is what it added beyond
  the slowest processor — zero for a pipeline whose services take longer to
  connect than warming takes to load. Observers can also handle the new
  `BaseObserver.on_startup_warmup()` event directly.
  (PR [#5600](https://github.com/pipecat-ai/pipecat/pull/5600))

- `StartupTimingReport` now accounts for the whole of a pipeline's startup, so
  a slow start can be read off the report rather than inferred.
  `setup_phase_secs` and `start_phase_secs` split the span into the two phases
  that compose differently: processors are set up concurrently, so the longest
  single piece of work decides that phase, while the `StartFrame` reaches them
  one at a time, so what each spends on it adds up.
  `ProcessorStartupTiming.start_duration_secs` reports the latter per processor
  alongside the existing `setup_duration_secs`.
  (PR [#5600](https://github.com/pipecat-ai/pipecat/pull/5600))

- Added `language_codes` and `timestamps` settings to
  `AssemblyAISyncSTTService`, aligning it with the Sync API's config surface.
  `language_codes` (e.g. `[Language.EN, Language.ES]`) declares the audio's
  languages for multilingual or code-switching audio, and is bound in
  preference to the single `language` when both are set; regional variants
  resolve to their base code and duplicates are dropped, preserving declaration
  order. `timestamps` requests per-word `start`/`end` timings on the
  transcription result, and is unset by default so the API default of `False`
  applies.
  (PR [#5603](https://github.com/pipecat-ai/pipecat/pull/5603))

- Added `ServiceMetricsObserver`, which reports each metric a service publishes
  as a record rather than a log line. `on_service_latency` carries time to
  first byte, first audio and first answer token, with the parts each
  decomposes into; `on_service_usage` carries speech-to-text audio seconds,
  text-to-speech characters, and the token counts an LLM reports. One record
  per metric, never summed, so a consumer groups them by turn, session or
  model.
  (PR [#5607](https://github.com/pipecat-ai/pipecat/pull/5607))

- Added `SpeakingObserver`, which reports a conversation's speaking lifecycle
  through `on_speech_event`: `user_speech_started` / `user_speech_stopped` as
  the detector heard them, `user_turn_started` / `user_turn_stopped` as the
  turn strategy ruled on them, `bot_speech_started` / `bot_speech_stopped`, and
  `interruption`. Speech is timed to where it began and ended rather than to
  where the detector confirmed it, and a moment that closes a stretch of speech
  names its `started_at`, so an interval reads whole from one record.
  (PR [#5612](https://github.com/pipecat-ai/pipecat/pull/5612))

- Added `ErrorObserver`, which reports every error a pipeline raises through
  `on_error` as an `ErrorEvent`: the `message`, the `category` it was
  attributed to, the `exception_type` behind it, the `processor` that raised
  it, and whether that processor can still do its job (`processor_usable`).
  Errors are read where they are raised, so ones a processor answers for itself
  — a `ServiceSwitcher` failing over, or a service it holds in reserve — are
  reported too; `PipelineWorker`'s `on_pipeline_error` sees only the errors
  that reach the top of the pipeline.
  (PR [#5618](https://github.com/pipecat-ai/pipecat/pull/5618))

- Added `FunctionCallObserver`, which reports each function call through
  `on_function_call_event` as a `FunctionCallEvent`: `function_call_started`
  when the LLM asks for it, `function_call_in_progress` when it begins running,
  and one of `function_call_completed`, `function_call_failed`,
  `function_call_timed_out` or `function_call_cancelled` when it settles. Each
  moment names the one before it, so the wait to run and the time spent running
  each read from one record. Arguments travel by default and results do not;
  `include_arguments` and `include_results` change either.
  (PR [#5622](https://github.com/pipecat-ai/pipecat/pull/5622))

- `DeepgramFluxSTTService`, `DeepgramFluxSageMakerSTTService` and
  `CartesiaTurnsSTTService` push an `EagerTranscriptionFrame` when they predict
  an end of turn and an `EagerEndOfTurnCancelFrame` when they withdraw it,
  which `EagerUserTurnStrategies` acts on. Both frames are emitted only while
  `enable_eager_end_of_turn` is on.
  (PR [#5625](https://github.com/pipecat-ai/pipecat/pull/5625))

- Added support for answering an eager end of turn: a response is generated
  from a service's predicted end of turn and held until the turn is confirmed,
  so the gap before the committed end of turn is spent generating rather than
  waiting. Turn it on with `enable_eager_end_of_turn=True` on
  `DeepgramFluxSTTService`, `DeepgramFluxSageMakerSTTService` or
  `CartesiaTurnsSTTService`, which then recommend `EagerUserTurnStrategies` in
  place of `ExternalUserTurnStrategies`. The LLM service holds the response, so
  no extra processor is needed. The response is discarded if the user resumes
  speaking or if the committed transcript differs from the predicted one, per
  the strategy's `match_policy` (`NormalizedMatch` by default, which ignores
  the capitalization and punctuation services commonly add when committing a
  transcript, or `ExactMatch` to require the two to be identical). An
  unconfirmed turn never reaches the user or the LLM context, and tool calls
  are never executed for one.
  (PR [#5625](https://github.com/pipecat-ai/pipecat/pull/5625))

- Added `FlowConfig`, which describes a Pipecat Flows conversation as data so a
  bot can load its flow at runtime. A config names the initial node and, for
  each node, its messages, the tools it offers, and where each tool leads. Load
  it with `FlowConfig.from_file()`, `from_yaml()`, or `from_json()`, construct
  a `Flow` from it and the module holding your direct functions and action
  handlers, or a list of modules, and hand that to `FlowManager`.

    Tools stay in Python and return `(result, TRANSITION_IN_YAML)`; a function
  that only moves the conversation is written in the config alone, as a
  `transition_only` entry with a `description`. The config picks the next node,
  by name or with a branch table keyed on a field of the result, where case
  keys may be strings, booleans, or numbers. Constructing the `Flow` reports
  every tool, handler, or variable the config names but cannot resolve,
  together, as a `FlowReferenceError`. Actions use the built-in `tts_say`,
  `end_conversation`, and `function` types, or a custom type registered in code
  or given a `handler` name in the config. Prompts may use `{{ key }}`
  placeholders, which `FlowManager` fills from its `state` on entering the
  node, and a YAML config can keep long prompts in their own files with
  `!include path`, resolved relative to the config. The format's JSON Schema is
  kept in the repository at `src/pipecat/flows/flow_config.schema.json` for
  editors and other tools to vendor.

    ```yaml
    initial_node: greet
    nodes:
      greet:
        role_message: You take food orders for {{ restaurant_name }}.
        task_messages:
          - role: developer
            content: Greet the caller and ask: pizza or sushi?
        functions:
          - name: choose_pizza
            transition_only: true
            description: The caller wants pizza.
            transition_to: pizza
      pizza:
        task_messages:
          - role: developer
            content: Take the pizza order.
        functions:
          - name: select_pizza_order
            transition_to:
              field: status
              cases:
                ok: confirm
                unavailable: pizza
    ```

    ```python
    import handlers  # direct functions: select_pizza_order, ...; action
  handlers

    config = FlowConfig.from_file("flow.yaml")
    flow = Flow(config, handlers=handlers)
    flow_manager = FlowManager(
        worker=worker,
        llm=llm,
        context_aggregator=context_aggregator,
        global_functions=flow.global_functions,
    )
    flow_manager.state["restaurant_name"] = "Luigi's"  # fills {{
  restaurant_name }}
    await flow_manager.initialize(flow.initial_node)
    ```

    The flows examples are grouped by form: `examples/flows/yaml/` holds a
  folder per YAML flow with `bot.py`, `flow.yaml`, and `handlers.py`
  (hello_world, food_ordering, restaurant_reservation, patient_intake,
  insurance_quote, and podcast_interview), and `examples/flows/python/` holds
  the flows that show what needs code. The README says when to write a flow in
  each form.
  (PR [#5628](https://github.com/pipecat-ai/pipecat/pull/5628))

- Added `pipecat.utils.yaml.include_loader()`, which returns a
  `yaml.SafeLoader` subclass that resolves `!include <relative-path>` tags
  against a base directory. Flow configs and eval scenarios use it; any
  YAML-based configuration can.
  (PR [#5628](https://github.com/pipecat-ai/pipecat/pull/5628))

- A suite manifest lists scripted and simulated scenarios side by side under
  `scenarios:`, and `pipecat eval suite -k simulation` (or `-k script`) runs
  only one kind.
  (PR [#5646](https://github.com/pipecat-ai/pipecat/pull/5646))

- Simulations for `pipecat eval`: give a scenario a `persona:` and a `goal:`
  instead of scripted turns, and an LLM plays the caller, holding the whole
  conversation with your bot, in text or over synthesized speech, and hanging
  up once it has what it came for. A judge then reads the transcript, the bot's
  tool calls in place, and decides whether the bot did its job (`success:`) and
  how every reply scored on your `metrics:`, judged per turn (`criterion`, with
  `min_score` as the share of turns that must pass) or measured from the run
  (`measure: turns`, `duration`, `words`, or `latency` against a range, or
  `function_calls` against the list of calls the bot should make, `[]` for
  none). `pipecat eval run` plays one; a suite runs it `runs:` times and every
  run must pass.
    ```yaml
    persona: "Jamie, booking dinner for two tonight."
    goal: "Book a table for two at 6 PM, then end the call."
    success: "the bot confirmed a reservation for two at 6 PM"
    metrics:
      - name: politeness
        criterion: "the reply is courteous, never curt or dismissive"
        min_score: 1
      - measure: latency
        max_value: 5
    runs: 3
    ```
  (PR [#5646](https://github.com/pipecat-ai/pipecat/pull/5646))

- `RTVIClientTransport` and `RTVIClientSerializer` let a Pipecat pipeline join
  a bot as an RTVI client. The eval harness now runs on them: its STT, TTS,
  persona, and recording are stages of a pipeline of its own, and each suite
  run gets its own process.
  (PR [#5646](https://github.com/pipecat-ai/pipecat/pull/5646))

- Added `require_given()` to `pipecat.utils.types`. It narrows a settings value
  past `NOT_GIVEN` and `None` like `assert_given()`, and also rejects the empty
  string, raising `ValueError("<what> must be specified")` for settings a
  service cannot run without.
  (PR [#5670](https://github.com/pipecat-ai/pipecat/pull/5670))

- A simulation's results tell judge trouble from bot trouble: a turn the judge
  left unanswered carries `verdict: none`, a failed metric carries a
  `failure_kind` (`judge_no`, `judge_no_verdict`, `out_of_range`,
  `function_calls`), and a judge that gives no verdict on the goal makes the
  run an error with kind `judge_no_verdict`, kept out of the suite's pass rate.
  A failed `latency` measure's reason says what was timed, the first token in
  text mode or the first spoken sentence in audio mode.
  (PR [#5678](https://github.com/pipecat-ai/pipecat/pull/5678))

- A simulation ends as soon as it is going nowhere: `silence` when neither side
  does anything for `max_silence_s` (30 s by default), and an error the moment
  the harness's own pipeline fails, the persona LLM first among them, instead
  of waiting out `max_duration_s`.
  (PR [#5678](https://github.com/pipecat-ai/pipecat/pull/5678))

- A simulation's `simulator:` block is optional: without one the persona runs
  on the same local Ollama model as the default judge, `gemma4:12b`, so a
  simulation needs no API key. The release-eval simulations use it.
  (PR [#5678](https://github.com/pipecat-ai/pipecat/pull/5678))

- Added `BackendLLMWorker` and `BackendOutput` (`pipecat.workers.llm`): a
  worker that runs any LLM service, with its own context and multi-step tool
  calling, as the backend `OpenAILiveLLMService` client delegation hands work
  to. Everything the backend produces comes back as a `BackendOutput` saying
  what it is and whether the user may hear it, and `transform_output` decides
  that per output, or rewrites the text on its way out — see
  `examples/realtime/realtime-openai-live-client-delegation-speaking-policy.py`.
  (PR [#5688](https://github.com/pipecat-ai/pipecat/pull/5688))

- Added `OpenAILiveLLMService`, a speech-to-speech service for the OpenAI Live
  API (`gpt-live-1`). The live model is full-duplex — it listens and speaks at
  the same time and handles being interrupted itself — and delegates search,
  reasoning and tool use to a backend model while the conversation continues.
  Both delegation modes are supported:
  `OpenAILiveLLMService.ResponsesDelegation` lets OpenAI host the backend
  (Responses API) model, with its function calls executed by the pipeline's
  registered tool handlers; `OpenAILiveLLMService.ClientDelegation` runs any
  Pipecat LLM service as the backend through a `BackendLLMWorker`. See
  `examples/realtime/realtime-openai-live-responses-delegation.py` and
  `examples/realtime/realtime-openai-live-client-delegation.py`, plus
  `examples/realtime/realtime-openai-live-client-delegation-speaking-policy.py`
  for a backend that chooses what the user hears.
  (PR [#5688](https://github.com/pipecat-ai/pipecat/pull/5688))

### Changed

- ⚠️ **Behavior change:** `DeepgramSTTService` and
  `DeepgramSageMakerSTTService` no longer send `profanity_filter`, so
  Deepgram's own default of off applies. Deepgram's filter rewrites the words
  it matches rather than tagging them, so a false positive silently corrupts a
  transcript that may be stored, analyzed, or sent downstream. To keep the
  previous behavior, pass
  `settings=DeepgramSTTService.Settings(profanity_filter=True)`.
  (PR [#5462](https://github.com/pipecat-ai/pipecat/pull/5462))

- `CartesiaTTSService` and `CartesiaHttpTTSService` now default to `sonic-3.6`,
  Cartesia's current Sonic model. Pass `model="sonic-3.5"` to stay on the
  previous default.
  (PR [#5472](https://github.com/pipecat-ai/pipecat/pull/5472))

- `soundfile` is now a required dependency instead of an optional extra, so
  `SoundfileMixer` and the eval harness can load audio files with no extra
  install step. The `pipecat-ai[soundfile]` extra still resolves and is now a
  no-op, so existing installs keep working unchanged.
  (PR [#5473](https://github.com/pipecat-ai/pipecat/pull/5473))

- `text_contains` on a `user_transcription` eval expectation now accumulates
  the STT's final transcription segments within the turn before checking, so an
  STT that finalizes an utterance in pieces can still satisfy a phrase.
  Substring checks also ignore whitespace differences between the expected text
  and the event's text.
  (PR [#5473](https://github.com/pipecat-ai/pipecat/pull/5473))

- Raised the `pipecat-ai[cli]` Context Hub floor to `pipecat-ai-context-hub`
  0.6.0, which registers its MCP server with Claude Code for every directory
  rather than only the one `install` ran in. A project scaffolded after the hub
  was set up now has the server without re-running `install`.
  (PR [#5480](https://github.com/pipecat-ai/pipecat/pull/5480))

- Bots scaffolded by `pipecat init` build their `WorkerRunner` before wiring up
  event handlers, so handlers can end the session with `await runner.cancel()`.
  The runner takes `handle_sigint=runner_args.handle_sigint`, letting a bot run
  directly install its own signal handler while a hosted one leaves signals to
  its host.
  (PR [#5482](https://github.com/pipecat-ai/pipecat/pull/5482))

- The AssemblyAI STT examples and docs now name `universal-3-5-pro`,
  AssemblyAI's current Universal-3 Pro streaming model. The `u3-rt-pro` names
  are legacy aliases AssemblyAI redirects to it and are still accepted by
  `AssemblyAISTTService`.
  (PR [#5507](https://github.com/pipecat-ai/pipecat/pull/5507))

- `AWSTranscribeSTTService` now maps 46 more AWS Transcribe streaming
  languages, including Turkish, Hungarian, Tamil, Telugu, Swahili, Mexican
  Spanish and Welsh. Passing one of these as a `Language` enum previously sent
  a bare base code such as `tr`, which AWS Transcribe rejects.
  (PR [#5509](https://github.com/pipecat-ai/pipecat/pull/5509))

- `AWSPollyTTSService` now maps `Language.EN_IE` and `Language.EN_SG` to
  Polly's `en-IE` and `en-SG` locales, so Irish and Singaporean English no
  longer log an unverified-language warning.
  (PR [#5510](https://github.com/pipecat-ai/pipecat/pull/5510))

- Raised the `pipecat-ai[cli]` Context Hub floor to `pipecat-ai-context-hub`
  0.7.0, whose `refresh` indexes the framework at its newest release tag rather
  than tracking its default branch. Hub answers stop mixing in unreleased APIs
  stamped with the previous release's number, which `version_compatibility`
  reported as compatible to anyone running that release.

  Pass `--framework-version head` to index the default branch — the right
  choice when developing Pipecat itself or building against unreleased code. A
  `--framework-version` tag that does not exist now fails the refresh instead
  of logging a warning and exiting 0.
  (PR [#5513](https://github.com/pipecat-ai/pipecat/pull/5513))

- `SarvamTTSService` and `SarvamHttpTTSService` now default to `bulbul:v3`,
  Sarvam's current TTS model. Sarvam's API no longer serves `bulbul:v2` and
  rejects requests for it, so the previous default could not synthesize. The v3
  defaults come with it: `shubh` as the speaker and a 24000 Hz sample rate.
  (PR [#5517](https://github.com/pipecat-ai/pipecat/pull/5517))

- `SmallestTTSService` now maps all 31 languages the Waves Lightning API
  accepts, and no longer maps `Language.HE`, which the API rejects.
  (PR [#5519](https://github.com/pipecat-ai/pipecat/pull/5519))

- `FireworksLLMService` now defaults to
  `accounts/fireworks/models/nemotron-3-ultra-nvfp4`. Fireworks no longer
  serves the previous default, `accounts/fireworks/models/firefunction-v2`, on
  its serverless endpoint, so requests using it returned a 404.
  (PR [#5524](https://github.com/pipecat-ai/pipecat/pull/5524))

- `CartesiaTTSService` and `CartesiaHttpTTSService` now map `Language.OR`
  (Odia) and `Language.UR` (Urdu), the two languages Sonic 3.6 added.
  (PR [#5532](https://github.com/pipecat-ai/pipecat/pull/5532))

- Added a `turn_coverage` setting to `GeminiLiveLLMService.Settings`, which
  selects how much of the realtime input stream a user turn covers (for example
  `TURN_INCLUDES_ONLY_ACTIVITY` to keep video frames outside detected audio
  activity out of the turn).
  (PR [#5535](https://github.com/pipecat-ai/pipecat/pull/5535))

- `GroqTTSService` now supports a variety of sample rates, defaulting to 24khz.
  (PR [#5538](https://github.com/pipecat-ai/pipecat/pull/5538))

- `LiveAvatarNewSessionRequest` now defaults to `H264` video encoding, matching
  the LiveAvatar API's own default. LiveAvatar has deprecated `VP8`; pass
  `video_settings=VideoSettings(encoding=VideoEncoding.VP8)` to keep the
  previous encoding while it lasts.
  (PR [#5539](https://github.com/pipecat-ai/pipecat/pull/5539))

- `KokoroTTSService.Settings` has a `speed` field, kokoro-onnx's speech rate
  multiplier (0.5 to 2.0, default 1.0), settable at construction and at
  runtime.
  (PR [#5543](https://github.com/pipecat-ai/pipecat/pull/5543))

- `OpenAIResponsesLLMService`'s `service_tier` documentation now lists OpenAI's
  full set of tiers, including `fast` — the low-latency tier that replaced the
  `priority` name.
  (PR [#5548](https://github.com/pipecat-ai/pipecat/pull/5548))

- The OpenAI server-side turn detection example now uses `gpt-transcribe`,
  replacing `gpt-4o-transcribe`, which shuts down on 2027-02-26.
  (PR [#5549](https://github.com/pipecat-ai/pipecat/pull/5549))

- `OpenAISTTService` now defaults to `gpt-transcribe`, OpenAI's replacement for
  `gpt-4o-transcribe`, which shuts down on 2027-02-26. With
  `include_prob_metrics=True`, GPT transcription models now request `logprobs`
  alongside a `json` response; only Whisper models use `verbose_json`.
  (PR [#5549](https://github.com/pipecat-ai/pipecat/pull/5549))

- `PocketTTSService` now loads pocket-tts's distilled models for German,
  Spanish, Italian and Portuguese instead of the 24-layer variants, which are
  too slow to synthesize in real time on a typical CPU.
  (PR [#5553](https://github.com/pipecat-ai/pipecat/pull/5553))

- `FalImageGenService.Settings` now accepts `negative_prompt` and
  `guidance_scale`, passed through to Fal's image generation request when set.
  (PR [#5556](https://github.com/pipecat-ai/pipecat/pull/5556))

- `MoonshineSTTService` now maps `Language.DE` (German) and `Language.TL` /
  `Language.FIL` (Tagalog/Filipino), the languages Moonshine added to its model
  catalogue.
  (PR [#5557](https://github.com/pipecat-ai/pipecat/pull/5557))

- `TogetherLLMService` now defaults to `zai-org/GLM-5.2`. The previous default,
  `zai-org/GLM-5.1`, has been retired from Together's serverless inference and
  returns `model_not_available`. Pass
  `settings=TogetherLLMService.Settings(model=...)` to choose a different
  model.
  (PR [#5570](https://github.com/pipecat-ai/pipecat/pull/5570))

- `NeuphonicTTSService` and `NeuphonicHttpTTSService` now accept `temperature`
  in their settings, passing Neuphonic's synthesis randomness control (0.0–1.0)
  through to the API. Left unset, Neuphonic's own default applies.
  (PR [#5578](https://github.com/pipecat-ai/pipecat/pull/5578))

- `NvidiaLLMService` now defaults to `nvidia/nemotron-3-super-120b-a12b`. The
  previous default, `nvidia/nemotron-3-nano-30b-a3b`, reached end of life on
  NVIDIA's NIM cloud endpoint and now returns HTTP 410.a
  (PR [#5579](https://github.com/pipecat-ai/pipecat/pull/5579))

- ⚠️ **Behavior change:** A word-timestamp event that matches nothing left to
  speak is now dropped instead of being emitted as a `TTSTextFrame`. Passing it
  through put text the LLM never wrote into the conversation context. Nothing
  is lost: the text that word should have covered is carried by the word that
  puts the sentence back in step, or emitted when the audio context ends.
  (PR [#5585](https://github.com/pipecat-ai/pipecat/pull/5585))

- ⚠️ `Mem0MemoryService` now adds retrieved memories with the `"developer"`
  role instead of `"system"` when `add_as_system_message` is set. Memories are
  extra guidance for the model, not a system prompt. Services without a
  `"developer"` role receive them as `"user"` — the same treatment non-OpenAI
  services already gave the `"system"` message.
  (PR [#5596](https://github.com/pipecat-ai/pipecat/pull/5596))

- `SegmentedSTTService` now appends 0.5 s of silence to each speech segment
  before transcribing it, so the model hears the end of speech and finishes the
  last word instead of dropping or garbling it. This applies to every service
  built on it, such as `WhisperSTTService`, `OpenAISTTService`,
  `GroqSTTService` and `MoonshineSTTService`. The padding is submitted to the
  provider, so it counts toward STT usage metrics. Pass
  `trailing_silence_secs=0` to the service to disable it, or another value to
  tune it.
  (PR [#5621](https://github.com/pipecat-ai/pipecat/pull/5621))

- `FunctionCallResultFrame` now carries an `error` field, set when a call ends
  because its handler raised, so a failed call can be told apart from a
  successful one. The `result` is unchanged: it remains the stand-in message
  the LLM reads.
  (PR [#5622](https://github.com/pipecat-ai/pipecat/pull/5622))

- ⚠️ **Behavior change:** the Deepgram Flux services no longer push their eager
  end-of-turn transcript as an `InterimTranscriptionFrame`. That transcript is
  a prediction the service may withdraw, and it now travels as an
  `EagerTranscriptionFrame`, emitted only under `enable_eager_end_of_turn`, so
  it is no longer surfaced to clients as partial user speech. Flux emits no
  interim transcriptions as a result; subscribe to `on_update` for incremental
  transcript text.
  (PR [#5625](https://github.com/pipecat-ai/pipecat/pull/5625))

- The `on_user_turn_inference_triggered` event emitted by `UserTurnController`
  and by `BaseUserTurnStopStrategy` now carries a `UserTurnSpeculation | None`
  as its last argument, set when the inference answers a turn that hasn't ended
  yet and carrying the turn text it was run against. A custom stop strategy
  requests one by passing it to `trigger_user_turn_inference_triggered()`. The
  events of the same name on `LLMUserAggregator` and `UserTurnProcessor` are
  unchanged, so handlers registered on those keep their signature.

      ```python
      @controller.event_handler("on_user_turn_inference_triggered")
      async def on_user_turn_inference_triggered(controller, strategy,
  speculation): ...
      ```
  (PR [#5625](https://github.com/pipecat-ai/pipecat/pull/5625))

- `LLMContextFrame` carries a `speculation` flag marking an inference as
  speculative, so the response it produces is held until the turn it answers is
  confirmed.
  (PR [#5625](https://github.com/pipecat-ai/pipecat/pull/5625))

- The `AGENTS.md` that `pipecat init` writes now describes Pipecat Flows as
  part of Pipecat (`pipecat.flows`) and directs coding agents to build a new
  flow as a `FlowConfig` YAML file plus a Python tools module, reserving
  Python-built flows for routing or prompts that need code.
  (PR [#5628](https://github.com/pipecat-ai/pipecat/pull/5628))

- Eval suite `results.jsonl` records carry `scenario` and `kind` (`script` or
  `simulation`), and a repeated sweep (`--repeat`) reports pass rates and exits
  0.
  (PR [#5646](https://github.com/pipecat-ai/pipecat/pull/5646))

- `pipecat eval` loads the nearest `.env` (walking up from the working
  directory, shell variables winning) before running, so a simulation's persona
  LLM and a hosted judge find their credentials the way the bots do.
  `python-dotenv` is now a core dependency.
  (PR [#5646](https://github.com/pipecat-ai/pipecat/pull/5646))

- `CrusoeLLMService` now defaults to `openai/gpt-oss-120b`. The previous
  default was `zai/GLM-5.2`. Pass
  `settings=CrusoeLLMService.Settings(model=...)` to choose a different model.
  (PR [#5656](https://github.com/pipecat-ai/pipecat/pull/5656))

- The Speechify service now sends the standard `Speechify-Caller-Version`
  attribution header (valued from `pipecat_version()`), replacing the
  non-standard `X-Pipecat-Version` header.
  (PR [#5668](https://github.com/pipecat-ai/pipecat/pull/5668))

- The `runner` extra now requires `pipecat-ai-prebuilt>=1.1.0`. The prebuilt
  client UI served by the development runner gains a LiveKit option in its
  transport selector, backed by `@pipecat-ai/livekit-transport` 1.0.0, and
  moves to `@pipecat-ai/voice-ui-kit` 0.14.0.
  (PR [#5675](https://github.com/pipecat-ai/pipecat/pull/5675))

- `EvalSession.from_scenario()` builds the session for a loaded scenario of
  either kind, an `EvalScriptSession` for a scripted scenario or an
  `EvalSimulationSession` for a simulation, and `EvalSession` is the base class
  of both. Constructing `EvalSession(...)` directly is no longer supported; use
  the kind's own class.
  (PR [#5676](https://github.com/pipecat-ai/pipecat/pull/5676))

- `EvalClientParams` holds only what the scenario asks of the bot, and
  `EvalClient` takes it together with the run's `EvalSessionParams` and the
  services. `EvalClient.for_scenario()` and `EvalClient.for_simulation()` are
  gone; the sessions construct the client.
  (PR [#5676](https://github.com/pipecat-ai/pipecat/pull/5676))

- The eval sessions and `EvalSuite.run()` take how a run behaves as one
  `EvalSessionParams` (`connect_timeout_s`, `default_timeout_ms`,
  `record_path`, `cache_dir`, `use_cache`, `stop_bot`, `trigger_disconnect`),
  and the services a run uses (`judge`, `user_tts`, `bot_stt`, and a
  simulation's `persona_llm`) as keyword arguments on both `from_scenario()`
  and the constructors.
  (PR [#5676](https://github.com/pipecat-ai/pipecat/pull/5676))

- Cartesia word timestamps for Chinese and Japanese now keep one token per
  reported entry, each with its own start time, instead of joining a whole
  `word_timestamps` message into a single token.
  (PR [#5679](https://github.com/pipecat-ai/pipecat/pull/5679))

- `PiperTTSService` downloads voice models to `~/.cache/pipecat/piper` by
  default, the same cache location `KokoroTTSService` uses, instead of the
  current working directory. Pass `download_dir` to keep models elsewhere; a
  model already downloaded into a working directory is fetched again into the
  cache on first use.
  (PR [#5687](https://github.com/pipecat-ai/pipecat/pull/5687))

### Deprecated

- Deprecated `LatencyBreakdown.chronological_events()`. Use
  `LatencyBreakdown.turn_contribution_lines()` instead, which names every part
  of the interval rather than the services that happened to report a metric,
  and whose lines sum to the measured latency. Will be removed in 2.0.0.
  (PR [#5445](https://github.com/pipecat-ai/pipecat/pull/5445))

- Deprecated `SarvamTTSModel.BULBUL_V2`. Sarvam's API rejects `bulbul:v2` on
  both the REST and the WebSocket endpoint, so it cannot synthesize; use
  `SarvamTTSModel.BULBUL_V3`. Constructing `SarvamTTSService` or
  `SarvamHttpTTSService` with it still resolves a model config and now emits a
  `DeprecationWarning`. It will be removed in 2.0.0.
  (PR [#5517](https://github.com/pipecat-ai/pipecat/pull/5517))

- Deprecated the `on_progress` parameter of `EvalSession()` and
  `EvalSession.from_scenario()`, and the `on_update` parameter of
  `EvalSuite.run()`. Use the event handlers of the same name instead. Passing a
  callback still works, keeps its existing signature, and emits a
  `DeprecationWarning`. They will be removed in 2.0.0.
  (PR [#5520](https://github.com/pipecat-ai/pipecat/pull/5520))

- Setting the system prompt as a `"system"` message at the start of
  `LLMContext` is deprecated and will stop working in 2.0.0. Set it on the LLM
  service instead:

      ```python
      # Before
      context = LLMContext([{"role": "system", "content": "Be helpful."}])

      # After
      llm = OpenAILLMService(api_key=..., system_instruction="Be helpful.")
      context = LLMContext()
      ```

  To change the prompt at runtime, push an `LLMUpdateSettingsFrame`
  carrying `LLMSettings(system_instruction=...)` instead of rewriting the first
  context message.
  (PR [#5596](https://github.com/pipecat-ai/pipecat/pull/5596))

- `EvalScenario`, `EvalResult`, `EvalTurn`, `EvalTurnResult`, and
  `EvalTurnProgress` are now `EvalScriptScenario`, `EvalScriptResult`,
  `EvalScriptTurn`, `EvalScriptTurnResult`, and `EvalScriptTurnProgress`, the
  scripted kind of scenario. The old names remain as aliases until 2.0.0.
  (PR [#5646](https://github.com/pipecat-ai/pipecat/pull/5646))

- `pipecat.evals.harness` moved to `pipecat.evals.script_session`. The old
  module path remains until 2.0.0.
  (PR [#5646](https://github.com/pipecat-ai/pipecat/pull/5646))

- `RTVIEvalSerializer` is now `EvalSerializer`, the serializer of the bot's
  `EvalTransport`; its client-side counterpart is `EvalClientSerializer`. The
  old name remains as an alias until 2.0.0.
  (PR [#5646](https://github.com/pipecat-ai/pipecat/pull/5646))

- Deprecated the `connect_timeout_s`, `default_timeout_ms`, `record_path`,
  `cache_dir`, `use_cache`, `stop_bot`, and `trigger_disconnect` keyword
  arguments of `EvalSession.from_scenario()` and
  `EvalScriptSession.from_scenario()`, and the `use_cache` and
  `default_timeout_ms` keyword arguments of `EvalSuite.run()`. Pass
  `params=EvalSessionParams(...)` instead. The old arguments still work,
  override the `params` field of the same name, and emit a
  `DeprecationWarning`. They will be removed in 2.0.0.
  `EvalSimulationSession.from_scenario()` is new and takes `params` only.
  (PR [#5676](https://github.com/pipecat-ai/pipecat/pull/5676))

- Deprecated `service: openai` in a scenario's `judge.eval:` and `simulator:`
  blocks and `service: cartesia` in its `user.speech:` block, along with the
  `openai_service()` and `cartesia_service()` builders behind them. The
  built-in names are the local, keyless services (Ollama, Kokoro, Moonshine,
  Whisper); any other provider is a `factory:`, a dotted path to a callable
  that takes the block and returns the service, documented with examples in
  `pipecat.evals.services`. The old names still work and emit a
  `DeprecationWarning`; they will be removed in 2.0.0.
  (PR [#5678](https://github.com/pipecat-ai/pipecat/pull/5678))

### Removed

- Removed `SEND_CHUNK_MS` from `pipecat.evals.harness`. The harness now sends
  the user's audio through its own output transport, which paces it like any
  Pipecat transport does.\n
  (PR [#5646](https://github.com/pipecat-ai/pipecat/pull/5646))

### Fixed

- Fixed `AsyncAITTSService` ignoring runtime `model`, `voice` and `language`
  updates. Those three are sent only in the websocket init message and never
  repeated per utterance, so a change to any of them now opens a new session
  instead of being stored and warned about.
  (PR [#5464](https://github.com/pipecat-ai/pipecat/pull/5464))

- Deepgram Flux STT now applies a `model` or `numerals` update by reconnecting,
  since Flux reads those only from the connection URL. The reconnect waits
  until the user stops speaking, and settings Flux accepts on a live connection
  (`keyterm`, `eot_threshold`, `eager_eot_threshold`, `eot_timeout_ms`,
  `language_hints`) are still sent without dropping it.
  (PR [#5477](https://github.com/pipecat-ai/pipecat/pull/5477))

- A Deepgram Flux `FatalError` now reports the code and description Flux sends,
  instead of a generic `Unknown error`.
  (PR [#5477](https://github.com/pipecat-ai/pipecat/pull/5477))

- Deepgram Flux STT now fails instead of hanging when a connection cannot be
  established. A connection setting the endpoint rejects leaves the SageMaker
  session opening forever and produces no confirmation on either transport, so
  both the session handshake and the wait for Flux to confirm it are now
  bounded and reported through `on_connection_error`.
  (PR [#5498](https://github.com/pipecat-ai/pipecat/pull/5498))

- Deepgram Flux STT now marks itself unusable once it can no longer transcribe
  — after a fatal error from Flux, or a connection that never establishes — so
  the failure reaches the pipeline instead of the service staying connected and
  silent.
  (PR [#5498](https://github.com/pipecat-ai/pipecat/pull/5498))

- Fixed `GladiaSTTService` dropping Gladia's `stop_recording` message on a
  graceful end. The websocket closed before the message was sent, so sessions
  ended by dropping the connection rather than closing cleanly, and the
  service's `on_disconnected` event fired twice.
  (PR [#5499](https://github.com/pipecat-ai/pipecat/pull/5499))

- Fixed the pipeline idle timeout cancelling a live session while the user is
  still talking, on pipelines that get their turns from a provider instead of
  local VAD. `idle_timeout_frames` now counts `UserStartedSpeakingFrame`,
  `TranscriptionFrame` and `InterimTranscriptionFrame` as user activity,
  alongside `BotSpeakingFrame` and the VAD-driven `UserSpeakingFrame`.
  (PR [#5502](https://github.com/pipecat-ai/pipecat/pull/5502))

- Corrected the Nova Sonic region lists in `AWSNovaSonicLLMService`'s docstring
  and example: `eu-north-1` is a supported region for both Nova 2 Sonic and
  Nova Sonic.
  (PR [#5508](https://github.com/pipecat-ai/pipecat/pull/5508))

- Fixed the Vonage-Google package conflict so it no longer prevents
  `GeminiSTTService` usage in environments where the Vonage connector can't be
  installed anyway, such as development environments on macOS.
    The repo lockfile resolved `google-genai` to 2.8.0 — below the 2.9.0 the
  service needs — for every Python 3.13 environment, to satisfy
  `vonage-video-connector`'s `pydantic<2.12` pin, though the connector is
  installable only on Python 3.13 Linux. That environment is now resolved apart
  from every other. Installs from PyPI were never affected.
  (PR [#5512](https://github.com/pipecat-ai/pipecat/pull/5512))

- Fixed `SarvamHttpTTSService` ignoring its `sample_rate`. It named the field
  `sample_rate` in the request body where Sarvam's REST API reads
  `speech_sample_rate`, so synthesis always came back at the model's default
  rate while the audio frames were tagged with the requested one — audio played
  back at the wrong speed for any other rate.
  (PR [#5517](https://github.com/pipecat-ai/pipecat/pull/5517))

- Fixed the Fireworks function-calling example, which named
  `accounts/fireworks/models/gpt-oss-20b`. Fireworks retired that model from
  its serverless endpoint, so the example now uses
  `accounts/fireworks/models/nemotron-3-ultra-nvfp4`.
  (PR [#5524](https://github.com/pipecat-ai/pipecat/pull/5524))

- Fixed the `examples/flows/llm_switching.py` Bedrock service, which pinned a
  Claude model Bedrock has retired and a latency mode Bedrock does not offer
  for its replacement. It now uses
  `us.anthropic.claude-haiku-4-5-20251001-v1:0`.
  (PR [#5525](https://github.com/pipecat-ai/pipecat/pull/5525))

- Corrected `GeminiTTSService`'s documented model names: they differ between
  the Cloud Text-to-Speech and Gemini API backends, and the docstrings named
  only Cloud Text-to-Speech's `gemini-2.5-flash-tts` / `gemini-2.5-pro-tts`,
  which the Gemini API rejects with a 404.
  (PR [#5526](https://github.com/pipecat-ai/pipecat/pull/5526))

- Fixed Deepgram STT settings passed through `extra` being discarded when the
  key matched a declared settings field (for example `extra={"diarize":
  True}`). Such keys are now promoted to their field and sent to Deepgram.
  (PR [#5527](https://github.com/pipecat-ai/pipecat/pull/5527))

- Fixed `GeminiTTSService` ignoring its `language` setting on the Gemini API
  (GenAI) backend. The configured language is now sent as the speech config's
  `language_code`, as it already is on the Cloud Text-to-Speech backend.
  (PR [#5537](https://github.com/pipecat-ai/pipecat/pull/5537))

- `LmntTTSService` now maps all 31 languages LMNT's Blizzard model speaks,
  adding Assamese, Bengali, Czech, Danish, Finnish, Malayalam, Marathi, Slovak,
  Tamil and Telugu. These languages already reached the API through the
  base-code fallback, which logged a "not verified" warning for each of them.
  (PR [#5544](https://github.com/pipecat-ai/pipecat/pull/5544))

- Fixed an `OpenAISTTService` error where `include_prob_metrics=True` sent a
  request that diarization models reject. Those models report no per-token or
  per-segment probabilities, so the service now transcribes without requesting
  them and warns once.
  (PR [#5549](https://github.com/pipecat-ai/pipecat/pull/5549))

- Fixed the `examples/voice/voice-pockettts.py` import of `PocketTTSService`,
  which named a module that does not exist.
  (PR [#5553](https://github.com/pipecat-ai/pipecat/pull/5553))

- Fixed an issue where `ElevenLabsRealtimeSTTService` ignored the realtime
  API's `quota_exceeded`, `unaccepted_terms` and `invalid_request` messages
  instead of reporting them as errors.
  (PR [#5555](https://github.com/pipecat-ai/pipecat/pull/5555))

- Fixed the PCM sample rates `GrokRealtimeLLMService` accepts: the Grok Voice
  Agent API's 22050 Hz option is now allowed, and 21050 Hz — which the API
  rejects — is not.
  (PR [#5560](https://github.com/pipecat-ai/pipecat/pull/5560))

- Fixed the Rime TTS services' language mapping: Arabic, Italian, Japanese and
  Portuguese — all served by Rime's `coda` model — now resolve to Rime's
  language codes, and a region-qualified language such as `Language.EN_US`
  falls back to a base code Rime accepts instead of a BCP-47 tag that made Rime
  close the connection without synthesizing.
  (PR [#5564](https://github.com/pipecat-ai/pipecat/pull/5564))

- `SambaNovaLLMService` now sends the `frequency_penalty`, `presence_penalty`
  and `seed` settings, which SambaNova's chat completions API accepts. They
  were previously dropped from the request.
  (PR [#5565](https://github.com/pipecat-ai/pipecat/pull/5565))

- Fixed a `SpeechmaticsSTTService` error where setting `split_sentences` raised
  `ValueError: "VoiceAgentConfig" object has no field "split_sentences"` and
  the service failed to construct.
  (PR [#5568](https://github.com/pipecat-ai/pipecat/pull/5568))

- Fixed `SpeechmaticsTTSService` reporting the `speechmatics-rt` SDK version,
  rather than Pipecat's, in the `sm-app` tag it sends to Speechmatics. The tag
  now carries the Pipecat version, matching `SpeechmaticsSTTService`.
  (PR [#5569](https://github.com/pipecat-ai/pipecat/pull/5569))

- Fixed the NVIDIA examples `update-settings/llm/llm-nvidia.py` and
  `voice/voice-nvidia-sagemaker.py`, which requested Llama models that NVIDIA's
  NIM cloud endpoint no longer serves.
  (PR [#5579](https://github.com/pipecat-ai/pipecat/pull/5579))

- Corrected the `model` docstring in `OneShotInputParams` for
  `UltravoxRealtimeLLMService`: the field defaults to `None`, which lets
  Ultravox pick its current default model, rather than to the legacy
  `fixie-ai/ultravox` alias.
  (PR [#5582](https://github.com/pipecat-ai/pipecat/pull/5582))

- Fixed word-level tracking stalling for the rest of a sentence when a TTS
  service reports a word-timestamp event that doesn't match what is being
  spoken. The next correctly reported word is now matched past the gap and
  carries the text that was never reported, so `TTSTextFrame`s and RTVI
  progress keep pace with the audio. Previously the sentence stayed stuck at
  the first unmatched word, every word after it went unmatched too, and the
  remainder arrived as a single frame once the audio context ended.
  (PR [#5585](https://github.com/pipecat-ai/pipecat/pull/5585))

- Fixed `NvidiaLLMService` emitting reasoning frames before TTFB metrics, so
  observers receive TTFB before reasoning output for structured reasoning
  fields and inline `<think>` blocks.
  (PR [#5589](https://github.com/pipecat-ai/pipecat/pull/5589))

- The LLM usage debug log now includes cache creation tokens alongside cache
  reads.
  (PR [#5602](https://github.com/pipecat-ai/pipecat/pull/5602))

- OpenAI-compatible LLM services now report prompt-cache write tokens as
  `LLMTokenUsage.cache_creation_input_tokens`, so usage-based cost tracking
  accounts for them. Covers `BaseOpenAILLMService` and the services inheriting
  it, `OpenAIResponsesLLMService`, and `SambaNovaLLMService`. Providers that do
  not report the count leave the field unset.
  (PR [#5602](https://github.com/pipecat-ai/pipecat/pull/5602))

- `OutputAudioRawFrame`s pushed from upstream of an STT service (for example a
  sound effect or a recording played into the pipeline before the STT) now
  reach the output transport untouched. STT services only transcribe
  `InputAudioRawFrame`s. Previously any audio frame was treated as user speech,
  so the audio was sent to the transcription provider and, with
  `audio_passthrough=False`, never reached the transport.
  (PR [#5609](https://github.com/pipecat-ai/pipecat/pull/5609))

- `VoicemailDetector` no longer swallows `EndWorkerFrame` travelling upstream
  through a closed gate, so a handler that ends the call reaches
  `PipelineWorker`. Closed classifier and conversation gates already admitted
  `CancelWorkerFrame` (a system frame) but dropped other worker lifecycle
  frames.
  (PR [#5614](https://github.com/pipecat-ai/pipecat/pull/5614))

- `SegmentedSTTService` (Whisper, Moonshine, and the other segmented STT
  services) now transcribes each speech segment in a background task.
  Previously the transcription blocked the service's input audio frames, so a
  VAD downstream could not analyze them until it finished. Transcripts are
  still pushed in segment order.
  (PR [#5630](https://github.com/pipecat-ai/pipecat/pull/5630))

- A punctuation mark reported as its own word-timestamp entry no longer shifts
  the text around it or records the mark twice. Trailing punctuation is
  attributed to the word before it, so the mark's own event arrived with
  nothing left to account for and spent a character of the following word
  instead — no text was lost and the cursors resynced a word later, but the
  words in between were recorded with the break misplaced (`"Yeah"`, `","`,
  `"I"` gave `"Yeah, I c an help"`), and a turn ending inside that window kept
  the misplaced split. Where a service forwards raw tokens
  (`pre_merge_tokens=False`) the mark was also written to the context a second
  time. Most visible for languages reported one character at a time, where
  every mark is affected, though not specific to them.
  (PR [#5633](https://github.com/pipecat-ai/pipecat/pull/5633))

- `LLMService` now logs the composed system instruction only when it changes.
  Previously every tool sync recomposed it, so the full prompt was logged once
  per turn.
  (PR [#5635](https://github.com/pipecat-ai/pipecat/pull/5635))

- Fixed `LLMContextSummarizer`'s message-count trigger
  (`max_unsummarized_messages`), which never reset after a summary was applied
  and counted the injected summary message as a new conversation turn.
  (PR [#5642](https://github.com/pipecat-ai/pipecat/pull/5642))

- Bots scaffolded by `pipecat init` with Sarvam, Whisper, or Moonshine STT,
  Sarvam TTS, or Google STT no longer crash at startup. Sarvam and Google STT
  now use each service's default model or location, and Whisper and Moonshine
  read an optional `WHISPER_MODEL` or `MOONSHINE_MODEL` variable that falls
  back to the default model. Previously the generated code read a variable the
  generated `.env.example` never listed, and the unset variable arrived as an
  explicit `None` that overrode the working default and failed validation.
  (PR [#5661](https://github.com/pipecat-ai/pipecat/pull/5661))

- The generated `.env.example` now lists every variable the scaffolded bot
  reads. This adds entries for ElevenLabs realtime STT, Gemini STT and TTS,
  Mistral STT, xAI STT, Soniox TTS, Deepgram Flux TTS, Deepgram Flux SageMaker
  TTS, and the NVIDIA TTS voice ID, names the Fal key `FAL_API_KEY` to match
  the generated code, and reads the AWS Transcribe session token from
  `AWS_SESSION_TOKEN` instead of the misspelled `AWS_AWS_SESSION_TOKEN`.
  (PR [#5661](https://github.com/pipecat-ai/pipecat/pull/5661))

- Fixed `WhisperSTTService` raising `ValueError` at construction when
  `Settings(language=None)` is paired with a multilingual model. `None` asks
  Whisper to detect the language itself, which every model supports; the
  language check previously looked `None` up in the model's supported-language
  list.
  (PR [#5669](https://github.com/pipecat-ai/pipecat/pull/5669))

- Fixed the `Model.DISTIL_LARGE_V2` and `MLXModel.DISTIL_LARGE_V3` Whisper
  models being documented as multilingual. Every Distil-Whisper release is
  English-only.
  (PR [#5669](https://github.com/pipecat-ai/pipecat/pull/5669))

- Local STT and TTS services now reject a missing or blank model or voice at
  startup with a "must be specified" error. Piper, Pocket TTS, Kokoro, and XTTS
  check the voice in the constructor (Pocket TTS before loading its model, and
  Kokoro also against the loaded voices file), and Whisper, Whisper MLX, and
  Moonshine check the model. Previously a blank value, which is what a copied
  `.env.example` produces, reached the underlying library and failed with an
  unrelated message, sometimes only on the first utterance. XTTS also reports
  an unknown studio speaker as an error frame instead of raising a `KeyError`.
  (PR [#5670](https://github.com/pipecat-ai/pipecat/pull/5670))

- Fixed `XAISTTService` emitting each utterance's text twice. xAI sends a chunk
  final for every locked segment and then an utterance final that restates the
  whole utterance, and both were pushed as `TranscriptionFrame`, so LLM context
  aggregators and other consumers that accumulate finals saw the text
  duplicated. Chunk finals are now pushed as `InterimTranscriptionFrame` and
  only the utterance final becomes a `TranscriptionFrame`, always with
  `finalized=True`.
  (PR [#5673](https://github.com/pipecat-ai/pipecat/pull/5673))

- Fixed `SonioxSTTService` dropping Soniox's end-of-audio frame on a graceful
  end. The websocket closed before it was sent, so sessions ended by dropping
  the connection rather than closing cleanly, and the service's
  `on_disconnected` event fired twice.
  (PR [#5682](https://github.com/pipecat-ai/pipecat/pull/5682))

- Fixed `SarvamSTTService` mislabeling every transcript as Hindi when no
  language was detected, the default behavior in `transcribe` mode. Unresolved
  languages now stay `None`, and `ur-IN`, `mai-IN`, `sd-IN`, and `kok-IN` are
  now correctly mapped.
  (PR [#5692](https://github.com/pipecat-ai/pipecat/pull/5692))

- The `pipecat eval suite` dashboard fits the terminal: rows are windowed to
  the active runs, including in a repeated sweep, and never wrap, so the tally
  at the bottom is always on screen. A running row's spinner animates and each
  row carries a clock, from its first attempt's start to its last attempt's
  end; the failures list is set apart from its header by a blank line.
  (PR [#5694](https://github.com/pipecat-ai/pipecat/pull/5694))

### Other

- `CartesiaSTTService`, `DeepgramFluxSTTService`, and
  `DeepgramFluxSageMakerSTTService` now log the provider's request ID when the
  connection is established, so a session can be traced in provider-side
  diagnostics.
  (PR [#5475](https://github.com/pipecat-ai/pipecat/pull/5475))

- Documented `SonioxSTTService`'s `context` setting as accepting either a
  `SonioxContextObject` or a plain string of background text on any current
  Soniox model, along with the 8k-token cap Soniox applies to the context
  object.
  (PR [#5566](https://github.com/pipecat-ai/pipecat/pull/5566))

## [1.8.1] - 2026-08-27

### Fixed

- A directory passed to `pipecat eval run` now reads `.yml` files as well as
  its `.yaml` ones, matching the scenario names a manifest resolves. A
  directory holding neither is still an error rather than an empty run.
  (PR [#5463](https://github.com/pipecat-ai/pipecat/pull/5463))

- Fixed `pipecat init` repeatedly offering to build a Context Hub index that
  already exists, and the stale-index warning never appearing, on any index
  refreshed by Context Hub v0.5.3 or later. A future hub release can no longer
  silence these checks.
  (PR [#5468](https://github.com/pipecat-ai/pipecat/pull/5468))

## [1.8.0] - 2026-08-26

### Added

- Added `image_url` support for Gemini adapter, enabling external URLs via
  `Part.from_uri()`.
  (PR [#3573](https://github.com/pipecat-ai/pipecat/pull/3573))

- Added Google Speech-to-Text v2 `adaptation` support to `GoogleSTTService`, so
  recognition can be biased toward domain terms using inline or referenced
  phrase sets. Configurable at construction and updatable at runtime.
  (PR [#4413](https://github.com/pipecat-ai/pipecat/pull/4413))

- Added `MCPClient(tools_arguments=...)`, which injects extra arguments into
  every call of a tool. Use it for arguments the model shouldn't choose — a
  fixed search mode, an account id, a caller-supplied filter. The pinned
  arguments override anything the model supplies, and are hidden from the
  schema it sees:

    ```python
    mcp = MCPClient(
        server_params=...,
        tools_arguments={"search": {"mode": "realtime"}},
    )
    ```

  In this example, the model only ever sees `search(query=...)`, while every
  call reaches the server as `search(query=..., mode="realtime")`.
  (PR [#4939](https://github.com/pipecat-ai/pipecat/pull/4939))

- Added `MCPClient.tools()`: `LLMContext(tools=await mcp.tools())` is now all
  you need to use MCP tools — connecting, tool registration, and closing the
  connection at pipeline end are automatic.
  (PR [#4939](https://github.com/pipecat-ai/pipecat/pull/4939))

- Added `KeenableWebSearch` (`pipecat.services.keenable.search`), an optional
  service that gives voice agents live web search and page reading via a hosted
  MCP server powered by [Keenable AI](https://keenable.ai). It exposes the
  server's `search_web_pages` (with optional site and date-range filters) and
  `fetch_page_content` tools — pass `await search.tools()` to your `LLMContext`
  and the tools register automatically (the connection is released
  automatically when the pipeline ends, too). Install with the `keenable`
  extra. Works keyless (`pro` mode); pass `api_key=` for higher rate limits and
  access to the lower-latency `realtime` mode (requires an account with
  realtime mode enabled), selected with `mode="realtime"`.
  (PR [#4942](https://github.com/pipecat-ai/pipecat/pull/4942))

- Added the Pipecat Context Hub to the `cli` extra, so `uv tool install
  "pipecat-ai[cli]"` provides `pipecat context-hub` (alias `pipecat ch`) with
  no separate install — the guides `pipecat init` writes tell coding agents to
  query the hub, so the CLI ships it. Costs about 195 MB on top of the extra;
  bot runtimes are unaffected, since `cli` stays optional so base images remain
  lean. The scaffolded agent guides teach `pipecat context-hub` as the primary
  way to query the hub, with `uvx pipecat-ai-context-hub` as the no-install
  fallback.
  (PR [#5122](https://github.com/pipecat-ai/pipecat/pull/5122))

- Added Context Hub setup to `pipecat init`. On the coding-agent path it
  registers the hub's MCP server with each coding agent CLI it finds, and says
  what came of it: Cursor, VS Code, and Zed are configured by hand, so it
  points at `pipecat context-hub install` to print the config block to paste,
  and a client that rejects the registration reports why. It then offers to
  build the local index when there isn't one — a few minutes and roughly 900
  MB, so it asks rather than assumes. The question only appears while no index
  exists, so it doesn't return once you have one. `pipecat init quickstart`
  skips setup entirely to stay a short path to a running bot, and
  `--no-context-hub` opts out anywhere.
  (PR [#5122](https://github.com/pipecat-ai/pipecat/pull/5122))

- Added a Pipecat Context Hub freshness notice to the CLI. When a local hub
  index exists and has gone stale, or was built for a different `pipecat-ai`
  minor than the project in the working directory, the CLI prints a one-line
  hint on stderr suggesting `pipecat context-hub refresh` — so a coding agent
  citing an API that has since changed is caught before the generated code is.
  The check reads the hub's published index metadata directly with the standard
  library, adding no dependency and no meaningful startup cost. It is silent
  when no index exists, compares only `major.minor` (ignoring patch and dev
  segments), stays quiet for editable pipecat checkouts, and can be switched
  off with `PIPECAT_HUB_CHECK=0`; the staleness threshold shares the hub's own
  `PIPECAT_HUB_STALE_AFTER_DAYS`.
  (PR [#5122](https://github.com/pipecat-ai/pipecat/pull/5122))

- Added `ProposedUserStartedSpeakingFrame` and
  `ProposedUserStoppedSpeakingFrame`, the way a service with its own turn
  detection tells the pipeline where it thinks a turn boundary falls.
  `ExternalUserTurnStrategies` resolve those proposals into
  `UserStartedSpeakingFrame` / `UserStoppedSpeakingFrame`, so the strategies
  are a single place that decides turns and can be subclassed to adjust the
  timing — previously a service that emitted turn frames took the reins
  entirely and left nothing to extend. See
  `examples/turn-management/turn-management-custom-external-turn-strategy.py`
  for a stop strategy that holds the turn open past the service's proposal so a
  trailing afterthought can reopen it.

  Every in-repo service with built-in turn detection now emits proposed turn
  frames rather than real turn frames: the AssemblyAI, Cartesia Ink-2, Deepgram
  Flux, Gladia, Sarvam, Soniox, Speechmatics, and OpenAI Realtime STT services,
  and the OpenAI, xAI, and Inworld realtime LLM services. Third-party services
  that emit `UserStartedSpeakingFrame` / `UserStoppedSpeakingFrame` directly
  keep working unchanged; switching them to the proposal frames hands
  interruption handling back to the pipeline.

  `OpenAIRealtimeSTTService` (the transcription-only service, not the
  speech-to-speech `OpenAIRealtimeLLMService`) and `SarvamSTTService` now also
  recommend `ExternalUserTurnStrategies` when their server-side VAD is enabled,
  matching the other turn-detecting services. Their turn frames were previously
  informational and nothing in the pipeline acted on them.
  (PR [#5156](https://github.com/pipecat-ai/pipecat/pull/5156))

- Enabled MoQ client mode, where the bot and the browser both dial a relay and
  rendezvous there instead of the bot serving its own socket. Since neither
  side needs a reachable address, this works when the bot is behind NAT.
   - Select it by naming a relay: `python bot.py -t moq --moq-connect
      https://cdn.moq.dev/anon`. Without `--moq-connect` the bot serves its own
      socket, as before.
    - Each client-mode session gets its own random namespace, so concurrent
      sessions on a shared relay don't collide. Pass `--moq-namespace` to pin a
      well-known room instead.
    - Added `MOQParams.response_path` and `MOQParams.request_path`, which set the
      bot's broadcast paths directly (the bot publishes its `response_path`,
      subscribes to the peer's `request_path`) instead of deriving them from
     `namespace` + `participant_id` / `peer_id`.
    - The namespace layer needs both peers to agree on a namespace up front.
      These are for deployments where the paths are assigned externally instead —
      e.g. a host that runs one bot per caller and names both paths after an id the
      caller minted, so there's no namespace to agree on.
    - Either can be set alone; the other still derives from the namespace.
      Unset, behaviour is unchanged.
    - The default participant ids are now named by direction: the bot publishes
      under `<namespace>/response` and subscribes to the peer at
      `<namespace>/request` (previously `bot0` / `client0`). `--moq-bot-id` /
      `--moq-client-id` still override them.
  (PR [#5158](https://github.com/pipecat-ai/pipecat/pull/5158))

- Added an optional `language` key to the eval harness's built-in
  `user.speech:` and `judge.transcription:` blocks. Each built-in speech
  service builder (`kokoro`, `cartesia`, `whisper`, `moonshine`) now forwards
  `language` (a code like `zh` or a `Language`) into the service settings, so
  non-English audio evals can synthesize user turns and transcribe bot audio in
  the right language without the `factory:` escape hatch. Omitting `language`
  is unchanged; the TTS audio cache key now includes the language so English
  and non-English renders of the same text don't collide.
  (PR [#5171](https://github.com/pipecat-ai/pipecat/pull/5171))

- Added `JobParams` and `JobGroupParams`, which carry everything a job dispatch
  needs in one object: `name`, `payload`, `timeout`, plus `cancel_on_error` for
  groups and `label` / `cancellable` for how the work presents to a client UI.
  Pass one to `job(...)`, `job_group(...)`, `request_job(...)`, or
  `request_job_group(...)`:

    ```python
    job_id = await ui_jobs.request_job_group(
        "wikipedia",
        "news",
        params=JobGroupParams(payload={"query": query}, label=f"Research:
  {query}"),
    )
    ```
  (PR [#5221](https://github.com/pipecat-ai/pipecat/pull/5221))

- Added `BaseUIWorker`, a worker that surfaces its jobs and job groups on the
  client UI without involving an LLM. Every group it dispatches streams its
  lifecycle to the client as the standard `ui-job-group` envelopes, with the
  client's reserved `__cancel_job_group` event honored for groups dispatched as
  cancellable. Dispatch from a plain `BaseWorker` when the work should stay
  invisible. It is instantiable directly, so an app can register one on the
  runner as a dispatcher and call it from a tool, and `UIWorker` now inherits
  from it, keeping the same capability for a page-driving LLM worker.
  `BaseWorker` itself is unchanged. The `async-tasks` example fans out research
  through a `BaseUIWorker` dispatcher driven by the main pipeline's own LLM
  tool, one LLM instead of two, while `document-review` keeps its `UIWorker`,
  which reads and drives the page content its review depends on.
  (PR [#5221](https://github.com/pipecat-ai/pipecat/pull/5221))

- Added `WorkerRunner.get_worker(name)`, which returns a worker added to that
  runner, along with `BaseWorker.worker_runner`,
  `FrameProcessor.worker_runner`, and `FunctionCallParams.worker_runner` to
  reach the runner from inside a worker, a processor, or a tool handler. A tool
  that needs a peer worker can now find it by name rather than having the
  application pass the object in through `app_resources`:

    ```python
    async def research(params: FunctionCallParams, query: str):
        ui_jobs = params.worker_runner.get_worker("ui-jobs")
    ```

  Only workers on the same runner have a local instance to return; a worker
  on another runner is addressable over the bus but has no object to hand back.
  (PR [#5221](https://github.com/pipecat-ai/pipecat/pull/5221))

- Added `BaseWorker.request_cancel_job_group(job_id, reason=...)`, the door for
  cancellation asked for from outside the worker. It honors the request only
  for a group dispatched with `JobGroupParams(cancellable=True)` and returns
  whether it did, so a client UI, an operator endpoint, or anything else
  reaching in gets the same rule. Cancellation the worker decides on itself, on
  shutdown, on a timeout, or through `cancel_on_error`, still calls
  `cancel_job_group()` and is never refused.
  (PR [#5221](https://github.com/pipecat-ai/pipecat/pull/5221))

- Added an `extra_headers` argument to `CartesiaSTTService`,
  `CartesiaTTSService` and `CartesiaHttpTTSService`, matching
  `CartesiaTurnsSTTService`. The headers are sent with the websocket handshake
  (or with each synthesis request, for `CartesiaHttpTTSService`), so
  deployments can supply their own authentication or routing headers.
  (PR [#5223](https://github.com/pipecat-ai/pipecat/pull/5223))

- Added `AudioVolumeTracker` (`pipecat.audio.volume`), which measures the
  volume of an audio stream over a rolling 400ms window. Audio is fed in chunks
  of any size with `update(audio, sample_rate)` and read back from the `volume`
  property, which reads 0 until the window holds enough audio to be measurable.
  Measuring happens on read and is cached until more audio arrives, so callers
  that report volume less often than they receive audio pay only for the reads.
  `VADAnalyzer` and `RTVIObserver` both track volume through it.
  (PR [#5232](https://github.com/pipecat-ai/pipecat/pull/5232))

- `AICFilter` and `AICQuailVADAnalyzer` now close their ai-coustics session
  when the pipeline stops, instead of waiting for garbage collection.
  (PR [#5239](https://github.com/pipecat-ai/pipecat/pull/5239))

- Added `PipelineWorker(processor_unusable_policy=...)`, deciding what the
  pipeline does when a processor reports an error that leaves it unable to do
  its job (becomes `is_usable=False`), such as a service whose API key was
  rejected.

  `ProcessorUnusablePolicy.CONTINUE` (the default) keeps the pipeline
  running and leaves the decision to the application, while `END` and `CANCEL`
  stop it gracefully or immediately. It is applied once per processor, not once
  per failed request:

      ```python
      worker = PipelineWorker(
          pipeline,
          processor_unusable_policy=ProcessorUnusablePolicy.END,
      )
      ```
  (PR [#5242](https://github.com/pipecat-ai/pipecat/pull/5242))

- Added `FrameProcessor.is_usable`, reporting whether a processor can still do
  its job, so applications can tell one that's briefly struggling from one that
  will never work again until something changes.

  A processor stays usable through failures it might recover from, and
  becomes unusable once its work can no longer succeed: a provider has rejected
  its API key, model or voice, or it has failed enough times to stop trying.
  Services stop accepting work and stop reconnecting once that happens, instead
  of retrying something that will keep failing.

  Errors set it as they are reported, so an error handler reading
  `frame.processor.is_usable` always sees the verdict that came with the error
  it is handling. This works in a worker's `on_pipeline_error` handler, which
  sees every error in the pipeline:

      ```python
      @worker.event_handler("on_pipeline_error")
      async def on_pipeline_error(worker, frame):
          if frame.processor and not frame.processor.is_usable:
              logger.error(f"{frame.processor} can no longer do its job:
  {frame.error}")
      ```

  and equally in a single processor's own `on_error` handler, when only one
  service is of interest:

      ```python
      @tts.event_handler("on_error")
      async def on_error(processor, frame):
          if not processor.is_usable:
              logger.error(f"TTS can no longer do its job: {frame.error}")
      ```

  Changes are also reported through `on_usable_changed`, a new event
  handler on the **processor**, which fires on the transition rather than on
  every error.

  Bring a processor back with `set_usable(True)` once whatever stopped it
  working has been dealt with. Services do this for themselves whenever their
  settings change, since a new model or voice may be exactly the fix.
  Credentials aren't runtime settings, so a rejected API key needs either a new
  service or an explicit `set_usable(True)`.
  (PR [#5242](https://github.com/pipecat-ai/pipecat/pull/5242))

- Added `ErrorCategory`, recording what kind of failure an error was — a
  rejected API key (`AUTHENTICATION`) versus a provider outage (`SERVER`), for
  example.

  `ErrorFrame` carries it in a new `category` field, and
  `FrameProcessor.push_error()` accepts it as an argument:

      ```python
      await self.push_error("rejected API key",
  category=ErrorCategory.AUTHENTICATION)
      ```

  The category says what went wrong, not what became of the processor. To
  decide whether a processor is worth using again, read `processor.is_usable`.

  Every error reaching a handler carries a category;
  `ErrorCategory.UNKNOWN` means the cause couldn't be determined, which
  handlers can treat the way they treated every error before.

  `FrameProcessor.push_error()` and `push_error_frame()` also take a
  `force_treat_as_permanent` argument, for an error that will keep recurring
  and so leaves the processor unable to do any more work. It's only needed for
  failures the category doesn't already convey, such as a websocket service
  exhausting its reconnection attempts; leaving it unset doesn't keep the
  processor usable, since a permanent category costs it its `is_usable` on its
  own.

  A category is worked out from the exception only when the reporter left
  it unset, so an error is never mistaken for a verdict on a processor it
  didn't come from:

    - Failures in application code a service invoked are reported as
      `ErrorCategory.APPLICATION`. A tool handler or TTS text transformer whose own
      API call returns 401 leaves the service usable, since its credentials were
      never in question.
    - Errors caught by a broad `except`, which may not have come from the
      processor at all, are reported as `ErrorCategory.UNKNOWN`.

  Classification falls back to the HTTP status code the exception carries.
  Processors whose provider signals failures through SDK-specific exceptions,
  or whose credentials can be rejected for reasons a reconnection would clear,
  refine it by overriding `_classify_error()`:

      ```python
      class MyService(TTSService):
          def _classify_error(self, exception: Exception) -> ErrorCategory |
  None:
              if isinstance(exception, MyProviderAuthError):
                  return ErrorCategory.AUTHENTICATION
              return None
      ```
  (PR [#5242](https://github.com/pipecat-ai/pipecat/pull/5242))

- A scenario's `judge:` block accepts an `extra:` mapping, forwarded to the
  judge model as top-level request parameters. This is how provider-specific
  options reach the judge; the default judge uses `reasoning_effort: none` so
  that a thinking-capable model does not spend latency, or the token budget its
  verdict needs, on reasoning that is never read.
  (PR [#5243](https://github.com/pipecat-ai/pipecat/pull/5243))

- `GoogleLLMService` now logs a warning naming Gemini's `finish_reason` when a
  response ends for a notable reason — withheld for safety or recitation, a
  rejected tool call, or truncated at the output token limit. Previously these
  ended the turn with little or no text and no indication why. Whatever text
  did arrive is still passed downstream, and responses ending normally are
  unaffected.
  (PR [#5248](https://github.com/pipecat-ai/pipecat/pull/5248))

- `GoogleLLMService` now bounds how long it waits for a streamed response, via
  a new `stream_idle_timeout_secs` argument that defaults to 20 seconds.
  Previously a stream that stopped producing without closing left the turn open
  indefinitely, since the API client applies no timeout of its own. Reaching
  the timeout fires `on_completion_timeout`, pushes an `ErrorFrame`, and closes
  the response, so the pipeline continues with whatever text arrived. The
  timeout covers the gap between chunks rather than the response as a whole,
  leaving a slow but healthy stream free to take as long as it needs. Raise it
  for models configured to think at length, since thinking emits no chunks, or
  pass `None` to wait indefinitely.
  (PR [#5249](https://github.com/pipecat-ai/pipecat/pull/5249))

- Added `FrameProcessor.pause_processing_all_frames_until(ready, timeout=...)`,
  which holds frames arriving at a processor until a condition resolves and
  then delivers them in order. Useful for a processor that establishes a
  connection in the background and cannot act on frames the moment it starts.

  `ready` is anything awaitable, typically an `asyncio.Event.wait` the
  processor already owns, so each service decides what "ready" means. The pause
  takes hold from the frame after the one being processed, so a `StartFrame`
  that triggers it still travels on downstream and pipeline startup is not
  delayed. Both frame queues are held, so `timeout` bounds the wait and the
  pause is always lifted, at the latest during cleanup.
  (PR [#5254](https://github.com/pipecat-ai/pipecat/pull/5254))

- Added full client/server coverage for xAI Voice Agent item truncate/delete,
  `force_message`, idle-timeout / DTMF / MCP event hooks, and session fields
  (`reasoning`, `resumption`, `replace`, transcription, VAD idle timeout) on
  `GrokRealtimeLLMService`.
  (PR [#5255](https://github.com/pipecat-ai/pipecat/pull/5255))

- Added Speechify to the text-to-speech services offered by `pipecat create`,
  which scaffolds a bot wired to `SpeechifyHttpTTSService` and adds
  `SPEECHIFY_API_KEY` and `SPEECHIFY_VOICE_ID` to the generated project's
  `env.example`.
  (PR [#5259](https://github.com/pipecat-ai/pipecat/pull/5259))

- Added `SpeechifyHttpTTSService`, a Speechify text-to-speech service backed by
  the `/v1/audio/stream/with-timestamps` endpoint. Audio and word-level speech
  marks arrive together over Server-Sent Events, so bot speech is attributed to
  the conversation context word by word and an interruption commits only the
  portion actually spoken. Speech marks require a streaming-native model: the
  service defaults to `simba-3.2` (English), and `simba-3.0` covers the other
  supported languages.

      ```python
      from pipecat.services.speechify.tts import SpeechifyHttpTTSService

      tts = SpeechifyHttpTTSService(
          api_key=os.environ["SPEECHIFY_API_KEY"],
          aiohttp_session=session,
          settings=SpeechifyHttpTTSService.Settings(voice="geffen_32"),
      )
      ```
  (PR [#5259](https://github.com/pipecat-ai/pipecat/pull/5259))

- Every eval suite run writes a `results.jsonl` next to its logs, one line per
  run with its outcome, its failures, and paths to its artifacts, appended as
  each run finishes so an interrupted sweep keeps everything already done. Runs
  that didn't pass also carry `events_seen`, the record of what the bot
  actually did. Each failure carries a machine-readable `kind` (`timeout`,
  `judge_no`, `missing_function_call`, ...; see `FAILURE_KINDS` in
  `pipecat.evals.harness`), which groups failures across many runs in a way the
  judge's free-text reasons cannot.
  (PR [#5260](https://github.com/pipecat-ai/pipecat/pull/5260))

- An eval suite can run each (bot, scenario) pair several times, via `repeat:`
  in the manifest or `--repeat N` on `pipecat eval suite`, and reports a pass
  rate per pair instead of a single verdict. This is how a behavior with a race
  in it — interruptions, async function results, turn detection — gets measured
  rather than sampled, since a bot that passes half the time looks identical to
  a reliable one in a single pass. Attempts interleave across bots (`A#1, B#1,
  C#1, A#2, ...`) so every bot meets the same machine conditions in the same
  stretch of the sweep, and each attempt's number joins its artifact filenames
  so nothing is overwritten. A repeated sweep always exits 0: it reports a
  rate, and what rate is acceptable is the caller's policy.
  (PR [#5260](https://github.com/pipecat-ai/pipecat/pull/5260))

- Added `retry_on_timeout` and `retry_timeout_secs` to `GoogleLLMService`,
  matching the OpenAI, Anthropic, and AWS services. With `retry_on_timeout`
  set, a request whose first chunk doesn't arrive within `retry_timeout_secs`
  is issued once more, so a request the API accepts and then never answers
  costs a few seconds instead of the whole idle timeout. Only the first chunk
  is retried, since re-issuing after that would duplicate the response.
  Gemini's client sends the request lazily, when the first chunk is pulled, so
  the window spans the whole round trip including any thinking the model does
  before it emits anything — leave it off for models that think at length.
  (PR [#5262](https://github.com/pipecat-ai/pipecat/pull/5262))

- Added `gemma4`, `glm5.2`, and `sarvam-105b-conversations` model support to
  `SarvamLLMService`. `gemma4` adds vision (inline data-URI image input),
  `glm5.2` adds reasoning support, and `sarvam-105b-conversations` targets
  multi-turn conversation on the `/v1` endpoint. The base URL is resolved
  automatically from the model (`/v1` for `sarvam-105b-conversations`, `/v2`
  for all others), and switching models at runtime recreates the client when
  the API version changes. Model-specific capabilities — vision,
  `reasoning_effort`, and `wiki_grounding` — are gated to the models that
  support them.
  (PR [#5288](https://github.com/pipecat-ai/pipecat/pull/5288))

- Added Icelandic, Sundanese, and Uzbek to the languages `SonioxTTSService` can
  speak.
  (PR [#5295](https://github.com/pipecat-ai/pipecat/pull/5295))

- `DeepgramFluxTTSService` now supports Flux's `speed` and `expressivity` voice
  controls, set via `DeepgramFluxTTSService.Settings` and updatable at runtime
  with a `TTSUpdateSettingsFrame`. A speed change is applied to the open
  connection with Flux's `Configure` message, so the cross-turn acoustic state
  survives it; expressivity is fixed when the connection opens, so a change
  reconnects. A settings update Deepgram rejects is reported as a non-fatal
  `ErrorFrame`.
  (PR [#5296](https://github.com/pipecat-ai/pipecat/pull/5296))

- Added LiveKit as a transport option in the development runner: `python bot.py
  -t livekit`, and `POST /start` support (`"transport": "livekit"`). Requires
  `LIVEKIT_URL`, `LIVEKIT_API_KEY`, and `LIVEKIT_API_SECRET` to be configured
  on the server.
  (PR [#5297](https://github.com/pipecat-ai/pipecat/pull/5297))

- Added `SarvamRealtimeSTTService` for low-latency streaming speech-to-text
  with Sarvam's `saaras:v3-realtime` model. Supports server-side endpointing
  (`endpointing="vad"`) and pipeline-driven endpointing
  (`endpointing="manual"`), interim and final transcripts, timestamps, and
  in-band configuration updates via `config.update`.
  (PR [#5301](https://github.com/pipecat-ai/pipecat/pull/5301))

- Added `cancellable_by_llm` to `@tool_options` and `register_function()`,
  which lets the LLM stop a running async tool call whose result the user no
  longer wants.

  A tool that opts in is advertised alongside its own `cancel_<name>`,
  which stops the one call of it that's running, and takes a `tool_call_id`
  only when several calls of that tool are running at once. A tool that doesn't
  opt in has no cancel tool and can't be stopped.

  Only applies when the `cancel_on_interruption=False` `@tool_options` is
  set. Consider using for long-running tool calls that a user might want to
  cancel, such as a long report or a background job that keeps producing
  results. The work has to outlast the LLM's route to cancelling it.

      ```python
      @tool_options(cancel_on_interruption=False, cancellable_by_llm=True)
      async def write_report(params: FunctionCallParams, topic: str):
          """Write a long research report on a topic.

          Args:
              topic: What the report should cover.
          """
          ...
      ```
  (PR [#5304](https://github.com/pipecat-ai/pipecat/pull/5304))

- Added OpenClaw Gateway support in `pipecat.services.openclaw`, for driving an
  OpenClaw coding agent from a pipeline.

  `OpenClawGatewayService` starts a run on an `OpenClawSendFrame`, redirects
  the one in flight on an `OpenClawSteerFrame`, and stops it on an
  `OpenClawAbortFrame`. A run answers with an `OpenClawStartedFrame`, any
  number of `OpenClawTextFrame`s, and one `OpenClawEndFrame` saying whether it
  completed, was cancelled, or failed. `OpenClawGatewayClient` speaks the same
  protocol without a pipeline.

  `examples/multi-worker/openclaw-agent` is a voice front end built on it.
  (PR [#5308](https://github.com/pipecat-ai/pipecat/pull/5308))

- Added the `function_call_stopped` scenario event to `pipecat.evals`, which
  reports a function call ending with the `tool_call_id` and a `cancelled` flag
  in its `args`. It takes the same `calls:` shape as `function_call`, so a
  scenario can assert how a call ended — telling work that was stopped from
  work that finished on its own, which a check on what the bot said about it
  cannot.

      ```yaml
      - event: function_call_stopped
        calls:
          - name: write_report
            args: { cancelled: true }
      ```
  (PR [#5314](https://github.com/pipecat-ai/pipecat/pull/5314))

- Added `setup_timeout_secs` and `start_timeout_secs` to `PipelineWorker`, both
  defaulting to 20 seconds. A processor that blocks while connecting, or while
  handling the `StartFrame`, would leave `run()` waiting on it forever; the
  pipeline is now torn down once the timeout elapses.
  (PR [#5316](https://github.com/pipecat-ai/pipecat/pull/5316))

- Added `acquires` and `releases` (`pipecat.utils.shared`) for a resource
  shared by several processors, such as the client an input and an output
  transport share. The first owner to acquire runs the decorated method while
  the rest wait for it, and only the last owner to release runs the undo. A
  method that raises is not attempted again: the exception reaches every owner,
  so the two halves of a transport either both come up or both fail.

      ```python
      class MyTransportClient:
          @acquires("client")
          async def setup(self, setup: FrameProcessorSetup): ...

          @releases("client")
          async def cleanup(self): ...
      ```
  (PR [#5316](https://github.com/pipecat-ai/pipecat/pull/5316))

- Added `BaseObserver.on_processor_setup`, called with a `ProcessorSetUp` once
  each processor has been set up. Services connect during setup, so this is
  where that cost can be measured; processors are set up concurrently, so these
  arrive in the order they finish rather than in pipeline order.
  (PR [#5316](https://github.com/pipecat-ai/pipecat/pull/5316))

- Added `on_setup_timeout` and `on_pipeline_timeout` events to
  `PipelineWorker`, so a pipeline that gives up waiting says so.
  `on_setup_timeout` fires when the processors never finish setting up, and
  takes no frame, since none has been pushed yet. `on_pipeline_timeout` fires
  when a frame the worker was waiting on never reaches the end of the pipeline:
  a `StartFrame` that never starts it, or a `CancelFrame` that never drains it,
  so inspect the frame to tell the two apart.

    ```python
    @worker.event_handler("on_pipeline_timeout")
    async def on_pipeline_timeout(worker, frame):
        if isinstance(frame, StartFrame):
            ...
    ```
  (PR [#5316](https://github.com/pipecat-ai/pipecat/pull/5316))

- Added a `stop_on_failure` field to eval scenarios. It defaults to `true`, the
  existing behavior, where the first turn with a failed assertion ends the
  scenario. Set it to `false` for a scenario whose turns are scored
  independently — a benchmark reporting a per-turn pass rate needs every turn
  driven, not just the ones before the first miss.
  (PR [#5317](https://github.com/pipecat-ai/pipecat/pull/5317))

- Added `TTFATMetricsData`, reporting time to first answer token for LLM
  services that answer in text. It runs from the request to the first token the
  caller sees, excluding any reasoning streamed first, and carries `ttfat`,
  `ttfb`, and `thinking_time` (`ttfat` minus `ttfb`) so the cost of a model
  thinking is readable from one metric. It reaches RTVI clients as `ttfat` and
  is logged by `MetricsLogObserver`. Speech-to-speech services report nothing,
  having no answer token to measure to.
  (PR [#5320](https://github.com/pipecat-ai/pipecat/pull/5320))

- Added `pcm_to_wav()` to `pipecat.audio.utils`, which wraps raw 16-bit PCM in
  a WAV container and returns the file as bytes. It takes PCM in the forms
  Pipecat pipelines carry it — `bytes`, `bytearray`, or `memoryview` — so audio
  from an `AudioBufferProcessor` event handler can be written or uploaded
  directly.
  (PR [#5326](https://github.com/pipecat-ai/pipecat/pull/5326))

- Added `AudioBufferProcessor` events `on_user_turn_audio` and
  `on_bot_turn_audio`, which fire once when a turn ends with a `TurnAudioData`
  holding that speaker's audio for the whole turn and the turn number. Turn
  tracking, which the pipeline worker enables by default, supplies the boundary
  and the number.
  (PR [#5329](https://github.com/pipecat-ai/pipecat/pull/5329))

- Added per-turn results to `pipecat.evals`. `EvalResult.turns` holds an
  `EvalTurnResult` for each turn in the scenario — its `status` (`passed`,
  `failed`, or `not_run`), the failures it produced, and its duration — so
  scoring a run turn by turn no longer means grouping `EvalResult.failures` by
  `turn_index` and opening the scenario file for a denominator. A turn the run
  stopped before reaching reports `not_run` instead of looking like a pass.
  `pipecat eval suite` writes the same statuses to each run's `results.jsonl`
  line, and `pipecat eval` prints a `2/4 turns` tally for a run that drove
  every turn and failed some.
  (PR [#5332](https://github.com/pipecat-ai/pipecat/pull/5332))

- Added `ElevenLabsDialogueTTSService`, a WebSocket TTS service for ElevenLabs
  Eleven v3 models (`eleven_v3` and `eleven_v3_conversational`), which
  `ElevenLabsTTSService` can't reach. It needs workspace access to ElevenLabs'
  Text-to-Dialogue API. `stability` is the only voice setting Text-to-Dialogue
  reads, and text is always aggregated into sentences
  (`TextAggregationMode.SENTENCES`):

    ```python
    from pipecat.services.elevenlabs.dialogue.tts import
  ElevenLabsDialogueTTSService

    tts = ElevenLabsDialogueTTSService(
        api_key=os.getenv("ELEVENLABS_API_KEY"),
        settings=ElevenLabsDialogueTTSService.Settings(
            voice=os.getenv("ELEVENLABS_VOICE_ID"),
            model="eleven_v3_conversational",
        ),
    )
    ```

  Keep using `ElevenLabsTTSService` for Flash, Turbo, and Multilingual
  models, which have lower latency and a fuller set of voice controls.
  (PR [#5353](https://github.com/pipecat-ai/pipecat/pull/5353))

- Added a warning when a `thinking_budget` is set on a Gemini 3 model. Gemini 3
  takes `thinking_level` instead. Passing `thinking_budget` results in
  ill-defined behavior (it may be honored, silently ignored, or rejected,
  depending on the model and the backend).
  (PR [#5356](https://github.com/pipecat-ai/pipecat/pull/5356))

- Added `DeepgramFluxSageMakerTTSService`, running Deepgram Flux TTS on a
  SageMaker endpoint. It takes `endpoint_name` and `region` instead of an API
  key and accepts the same settings as `DeepgramFluxTTSService`, including
  `voice`, `speed` and `expressivity`. Requires
  `pipecat-ai[deepgram,sagemaker]` and AWS credentials.
  (PR [#5360](https://github.com/pipecat-ai/pipecat/pull/5360))

- Added support for Azure's v1 API surface, which Microsoft Foundry displays as
  the "Azure OpenAI endpoint". `AzureLLMService` selects it whenever `endpoint`
  ends in `/openai/v1`, and `AzureRealtimeLLMService` reaches it from a
  `base_url` with no query string, appending the deployment named by
  `Settings.model`:

    ```python
    llm = AzureLLMService(
        api_key=os.getenv("AZURE_CHATGPT_API_KEY"),
        endpoint="https://my-resource.openai.azure.com/openai/v1",
        settings=AzureLLMService.Settings(model="my-deployment"),
    )

    realtime = AzureRealtimeLLMService(
        api_key=os.getenv("AZURE_REALTIME_API_KEY"),
        base_url="wss://my-resource.openai.azure.com/openai/v1/realtime",
        settings=AzureRealtimeLLMService.Settings(model="my-deployment"),
    )
    ```

  `AzureLLMService` still serves dated endpoints, routing them through
  `api_version`. For `AzureRealtimeLLMService`, v1 is the supported surface:
  endpoints carrying a dated `api-version` serve the superseded preview
  protocol, which rejects the session configuration and names its events
  differently, so they aren't usable here.
  (PR [#5363](https://github.com/pipecat-ai/pipecat/pull/5363))

- Added `token_provider` to `AzureLLMService` and `AzureRealtimeLLMService` for
  Microsoft Entra ID authentication, so Azure services can run without an API
  key. `api_key` is now optional, and passing neither credential raises
  `ValueError`:

    ```python
    from azure.identity.aio import DefaultAzureCredential,
  get_bearer_token_provider

    llm = AzureLLMService(
        token_provider=get_bearer_token_provider(
            DefaultAzureCredential(), "https://ai.azure.com/.default"
        ),
        endpoint="https://my-resource.openai.azure.com/openai/v1",
    )
    ```
  (PR [#5363](https://github.com/pipecat-ai/pipecat/pull/5363))

- Added `--ice-servers` to the development runner, along with the matching
  `PIPECAT_ICE_SERVERS` environment variable, so a bot started through
  `pipecat.runner.run.main()` can gather candidates from custom STUN and TURN
  servers:

    ```bash
    python bot.py -t webrtc --ice-servers stun:stun.l.google.com:19302
    ```

  An entry is a bare URL, or a JSON object with `urls`, `username`, and
  `credential` when a TURN server needs authentication. The environment
  variable takes the same entries comma-separated or as a JSON array.
  Configured servers also reach WebRTC clients in the `iceConfig` of the
  `/start` response, so both peers negotiate against the same servers.
  (PR [#5376](https://github.com/pipecat-ai/pipecat/pull/5376))

- Added `saaras:v4` to the models supported by `SarvamSTTService`. It uses the
  same WebSocket contract as `saaras:v3` — the same modes and fine-grained VAD
  tuning parameters — and adds Global English alongside Indian English and the
  22 Indic languages.
  (PR [#5382](https://github.com/pipecat-ai/pipecat/pull/5382))

- Added client-side TLS options to the MoQ transport:
  `client_tls_cert`/`client_tls_key` present a certificate to a relay that
  authenticates its peers with mTLS, and
  `client_tls_roots`/`client_tls_fingerprints` verify a relay behind a private
  CA or a self-signed one. The latter two are alternatives to switching
  `verify_ssl` off, which was previously the only way to reach such a relay; a
  bot in serve mode already publishes its own fingerprints as
  `MOQTransport.cert_fingerprints` for a peer to pin.
  (PR [#5387](https://github.com/pipecat-ai/pipecat/pull/5387))

- Added `BlandTTSService`, realtime WebSocket text-to-speech using Bland, and
  `BlandHttpTTSService` for complete-text HTTP requests. Install with `uv add
  "pipecat-ai[bland]"`.
  (PR [#5388](https://github.com/pipecat-ai/pipecat/pull/5388))

- Added `max_consecutive_zero_audio_contexts` to `TTSService`. A provider can
  accept every request and answer with silence — an unknown voice ID, say —
  without ever reporting an error, leaving the bot mute with nothing in the
  logs to explain it. Every TTS context that completes without producing audio
  reports an error the service can carry on from, so application code hears
  about a turn that produced no speech as it happens. After this many silent
  contexts in a row, the service reports a permanent error instead, stops being
  given work, and the pipeline worker applies its `ProcessorUnusablePolicy` (a
  `ServiceSwitcher` fails over to another provider). Defaults to 3; set it to 0
  to report silent contexts without ever writing the service off.
  (PR [#5393](https://github.com/pipecat-ai/pipecat/pull/5393))

- Added `PipelineWorker(handle_flush_frame=...)`, which says whether a worker
  answers a flush probe. It defaults to whether the pipeline is unbridged, so a
  bridged worker takes part in the trip but leaves the answering to the
  pipeline that owns the bridge, which is what makes `flush_pipeline()` on a
  bridged worker wait for what it produced to reach the end of that pipeline
  rather than only for its own queues to empty. A bridged worker with no such
  peer never completes a flush.
  (PR [#5399](https://github.com/pipecat-ai/pipecat/pull/5399))

- Added `BusSubscriber.accepts_bus_message(message)`, which the bus consults
  before every delivery to decide whether to hand the message to that
  subscriber. Returning `False` drops it for that subscriber alone; others
  still receive it. It accepts everything by default.
  (PR [#5399](https://github.com/pipecat-ai/pipecat/pull/5399))

- `pipecat eval run` now accepts a directory of `.yaml` scenario files and
  executes them in deterministic filename order.
  (PR [#5414](https://github.com/pipecat-ai/pipecat/pull/5414))

- Added `"max"` to the reasoning effort levels
  `OpenAIResponsesLLMService.ReasoningConfig` accepts, matching OpenAI's
  current set for the Responses API.
  (PR [#5432](https://github.com/pipecat-ai/pipecat/pull/5432))

- Added `complete_marker`, `incomplete_short_marker` and
  `incomplete_long_marker` to `UserTurnCompletionConfig`, so a bot can choose
  markers that are a single token in its own model's tokenizer. The turn
  completion instructions and both incomplete-turn re-prompts are rendered from
  whichever markers are configured.
  (PR [#5437](https://github.com/pipecat-ai/pipecat/pull/5437))

- Added `GeminiSTTService`, a streaming speech-to-text service using Google's
  `gemini-3.5-transcribe-live` model over the Gemini Live API. Language is
  auto-detected by default; `GeminiSTTService.Settings` supports `languages`
  hints and `adaptation_phrases` to bias recognition toward domain-specific
  terms. Requires google-genai >= 2.9.0.

  The model detects utterance boundaries itself, and when the pipeline's
  VAD signals end of speech the service flushes the utterance so the final
  transcript arrives promptly instead of when the model decides the utterance
  ended.
  (PR [#5449](https://github.com/pipecat-ai/pipecat/pull/5449))

### Changed

- ⚠️ `ExternalUserTurnStrategies`, when driven by the new proposed turn frames,
  now pushes `UserStartedSpeakingFrame` / `UserStoppedSpeakingFrame` and
  broadcasts the interruption itself rather than leaving both to the service.
  Fed real turn frames it still emits nothing, so pipelines built around a
  shared `UserTurnProcessor` or a third-party service that emits turn frames
  directly are unaffected.

  This matters if you pass `should_interrupt=False` to a turn-detecting STT
  *and* pin `user_turn_strategies=ExternalUserTurnStrategies()` by hand: the
  service carries `should_interrupt` on the strategies it recommends, but a
  user-supplied `user_turn_strategies` discards that recommendation, so
  interruptions come back on. Drop the manual `user_turn_strategies` — the
  service recommends the right strategies on its own now — or pass
  `ExternalUserTurnStrategies(enable_interruptions=False)`. The aggregator logs
  a warning naming both fixes when it detects this.

  Relatedly, a turn-detecting service paired with pinned non-external
  strategies (e.g. VAD or turn analyzer strategies) no longer drives turns at
  all; the pinned strategies own them.
  (PR [#5156](https://github.com/pipecat-ai/pipecat/pull/5156))

- Renamed `MOQParams.serve_bind` to `MOQParams.bind`, which now also sets the
  local source address a client-mode bot dials from.
  `MOQRunnerArguments.serve_bind` is renamed to match. The old name still works
  and warns; it will be removed in 2.0.0.
  (PR [#5158](https://github.com/pipecat-ai/pipecat/pull/5158))

- `MoonshineSTTService` now resolves languages the way the other STT services
  do: a `Language` maps to one of Moonshine's eight languages (Arabic, Chinese,
  English, Japanese, Korean, Spanish, Ukrainian, Vietnamese) through
  `language_to_moonshine_language()`, with regional variants such as
  `Language.ES_MX` resolving to their base code. A language Moonshine publishes
  no model for raises with the list of supported languages, instead of failing
  inside the model download.
  (PR [#5182](https://github.com/pipecat-ai/pipecat/pull/5182))

- `MoonshineSTTService` reloads its model when `language` or `model` changes at
  runtime (including via `set_language()`). Previously the new value was stored
  but the loaded model kept transcribing in the old language. A failed reload
  keeps the loaded model and pushes an `ErrorFrame`.
  (PR [#5182](https://github.com/pipecat-ai/pipecat/pull/5182))

- OpenAI-compatible LLM services now report token usage once per completion.
  Providers that repeat a cumulative usage snapshot on every streamed chunk
  previously produced a token-usage `MetricsFrame` for each one, over-counting
  a single turn for anything aggregating those frames. `SambaNovaLLMService`
  also now reports the cache-read and reasoning token counts its provider
  sends.
  (PR [#5190](https://github.com/pipecat-ai/pipecat/pull/5190))

- `GrokRealtimeLLMService`'s `voice` setting is typed `str` rather than a fixed
  list of five names, and accepts any built-in Grok voice ID (xAI documents the
  catalogue at https://docs.x.ai/docs/guides/voice/agent) or a custom ID from
  the Custom Voices API. Voice IDs are case-insensitive. `GrokVoice` is an
  alias of `str`.
  (PR [#5200](https://github.com/pipecat-ai/pipecat/pull/5200))

- **Behavior change:** the default Grok Realtime voice is now `eve`, the voice
  xAI documents as its default, instead of `Ara`. Anyone relying on the
  previous out-of-the-box voice should set `voice="ara"` explicitly on
  `SessionProperties`.
  (PR [#5200](https://github.com/pipecat-ai/pipecat/pull/5200))

- ⚠️ Streaming STT services no longer report processing metrics —
  `ProcessingMetricsData` in `MetricsFrame`, surfaced as the `processing` field
  of RTVI's `metrics` message. Nothing changes for `SegmentedSTTService`
  subclasses.

  Processing metrics time a discrete unit of work, and a streaming STT
  doesn't really perform one — audio arrives continuously. The 22 affected
  services' measurement methodologies were inconsistent and either not
  meaningful or duplicative of TTFB.

  TTFB — speech end to final transcript — is the STT latency measure, and it
  is unaffected.
  (PR [#5209](https://github.com/pipecat-ai/pipecat/pull/5209))

- ⚠️ `WebsocketTTSService` subclasses and `DeepgramSageMakerTTSService` no
  longer report processing metrics, which were meaninglessly reporting zero on
  every turn. The metric is `ProcessingMetricsData` in `MetricsFrame`, surfaced
  as the `processing` field of RTVI's `metrics` message. TTS services whose
  processing time was a real number are unaffected.

    Processing time is measured around `run_tts`. For a service that requests
  audio and waits for it in that call, that covers the real work.
  `WebsocketTTSService` subclasses instead push the text onto the socket and
  return, leaving the audio to arrive on a separate receive task, so the
  measurement only ever covered the send. `DeepgramSageMakerTTSService` does
  the same over bidirectional HTTP/2. TTFB and TTFA measure the latency that
  matters for all of them, and are unaffected.

    There's a new `TTSService.supports_processing_metrics` property, which
  defaults to `True`. Set it to `False` on a custom service whose `run_tts`
  returns before synthesis finishes, or back to `True` on a
  `WebsocketTTSService` subclass that waits for the server to signal the end.
  (PR [#5220](https://github.com/pipecat-ai/pipecat/pull/5220))

- Changed `JobGroup.worker_names` from a `set` to a `list`, preserving the
  order the workers were dispatched in so anything rendering them, such as a
  client UI job-group card, stays stable across a group's lifetime. `JobGroup`
  also now carries the group's `label` and `cancellable` settings and the set
  of workers that have reached a terminal state.
  (PR [#5221](https://github.com/pipecat-ai/pipecat/pull/5221))

- `CartesiaSTTService`'s `base_url` now also accepts a URL carrying a scheme
  (`ws://localhost:8000`) rather than only a bare host, so the connection can
  be made over plain `ws` against a local or proxied endpoint instead of always
  `wss`.
  (PR [#5223](https://github.com/pipecat-ai/pipecat/pull/5223))

- `CartesiaTTSService` now authenticates with the `X-API-Key` and
  `Cartesia-Version` headers on the websocket handshake instead of `api_key`
  and `cartesia_version` query parameters, matching the Cartesia STT services.
  (PR [#5223](https://github.com/pipecat-ai/pipecat/pull/5223))

- ⚠️ `calculate_audio_volume()` now requires at least 400ms of audio, the
  length of an ITU-R BS.1770 gating block, and raises `ValueError` for anything
  shorter. Code passing individual audio frames should use `AudioVolumeTracker`
  instead, which accumulates them into a rolling window. Volume is still
  reported on the same 0 to 1 scale, so `VADParams.min_volume` thresholds carry
  over unchanged.

  VAD continues to run on 32ms frames; only the volume measurement spans a
  wider window. Because loudness is now integrated over 400ms rather than a
  single frame, brief dips between phonemes no longer drop the measured volume
  below `min_volume` mid-word.
  (PR [#5232](https://github.com/pipecat-ai/pipecat/pull/5232))

- `GoogleLLMService` now defaults to `gemini-3.6-flash`, up from
  `gemini-2.5-flash`. 2.5 Flash follows the async-tool result-reporting
  instruction unreliably, and its failure mode is announcing a fabricated
  result rather than staying silent. Set `model` in `GoogleLLMService.Settings`
  to pin the previous default.
  (PR [#5236](https://github.com/pipecat-ai/pipecat/pull/5236))

- `AICQuailVADAnalyzer` now uses `vad-2.1-xxs-16khz` by default. The old
  default, `quail-vad-2.0-xxs-16khz`, does not work with `aic-sdk` 3.0. If you
  set `model_id` yourself, pick a model listed at
  https://artifacts.ai-coustics.io/.
  (PR [#5239](https://github.com/pipecat-ai/pipecat/pull/5239))

- The `aic` extra now requires `aic-sdk~=3.0`, which reworked its audio and VAD
  APIs. Upgrade the SDK when you upgrade pipecat; `aic-sdk` 2.5.x no longer
  works.
  (PR [#5239](https://github.com/pipecat-ai/pipecat/pull/5239))

- `ServiceSwitcher` now fails over only on errors that leave a service unable
  to do its job (`is_usable=False`), and reports its services' failures as its
  own.

  `ServiceSwitcherStrategyFailover` switches only once the active service
  reports an error that leaves it unable to do its job, rather than on any
  error, so a provider hiccup no longer costs a failover. It switches to the
  next service that is still usable, and a successful switch consumes the
  error: the switcher went on doing its job, so nothing upstream needs to act
  on it.

  The rest of the pipeline deals with the switcher rather than with the
  services inside it, so what it does with an error depends on which service
  reported it:

    - From a service it isn't using: the error stops at the switcher, since a
      service held in reserve can't stop the switcher doing its job. Watch that
      service's own `on_usable_changed` to hear about it.
    - From the active service, with somewhere to fail over to: consumed, as
      above.
    - From the active service, with nowhere left to go: re-reported against
      the switcher itself, naming the service that failed.

  The switcher's `is_usable` is a reading of its services: it reports
  itself unusable only once none of them can work, so one service's rejected
  API key never writes off the switcher along with it. Bringing any service
  back with `set_usable(True)` brings the switcher back with it; calling that
  on the switcher itself does nothing, since it has no usability of its own to
  set. The switcher raises `on_usable_changed` for itself whenever that reading
  moves, so watching the switcher is enough to hear about the services inside
  it.
  (PR [#5242](https://github.com/pipecat-ai/pipecat/pull/5242))

- Websocket services now stop reconnecting once the service can no longer do
  its job, instead of retrying credentials the provider has already rejected.

  Running out of reconnection attempts now leaves the service unusable too
  (`is_usable=False`), so a connection that can't be re-established is reported
  as such rather than being retried on every subsequent request. Errors
  reported during reconnection carry the exception that caused them, so they
  can be classified.

  Giving up is reported through the `report_error` callback, which takes an
  optional `force_treat_as_permanent` argument alongside the error frame. A
  service that overrides `_report_error` should accept and forward it.
  (PR [#5242](https://github.com/pipecat-ai/pipecat/pull/5242))

- STT and TTS services now stop working once they can no longer do their job —
  a bad API key, an unknown model or voice, a connection that won't come back —
  instead of retrying for every chunk of audio or piece of text.

  Previously a rejected API key on a service that connects on demand
  produced a connection attempt and an `ErrorFrame` several times a second for
  as long as the pipeline ran. `STTService` and `TTSService` now skip
  transcription and synthesis while the service is unusable.

  Services whose credentials are signed or resolved per connection —
  `AWSTranscribeSTTService` and `NvidiaSageMakerTTSService` — treat a rejected
  credential as recoverable, since reconnecting is what refreshes it.

  Pair this with `PipelineWorker(processor_unusable_policy=...)` or an
  `on_pipeline_error` handler to decide what the bot should do about it.
  (PR [#5242](https://github.com/pipecat-ai/pipecat/pull/5242))

- The eval judge now defaults to `gemma4:12b` with `reasoning_effort: none`,
  replacing `gemma2:9b`. Run `ollama pull gemma4:12b` before running scenarios
  that use the default judge. The previous default mistook a short interim
  reply for a complete answer — a bot that had so far said only "Let me check
  on that." would satisfy the criterion, passing a turn in which the bot said
  nothing. To keep the old judge, set it explicitly in a scenario's judge
  block: `judge: {eval: {service: ollama, model: gemma2:9b}}`.
  (PR [#5243](https://github.com/pipecat-ai/pipecat/pull/5243))

- Updated the default model for `DeepSeekLLMService` from `deepseek-chat` to
  `deepseek-v4-flash`. Set `model` in `DeepSeekLLMService.Settings` to pin the
  previous default.
  (PR [#5246](https://github.com/pipecat-ai/pipecat/pull/5246))

- Updated the default model for `MiniMaxHttpTTSService` from `speech-02-turbo`
  to `speech-2.8-turbo`. Set `model` in `MiniMaxHttpTTSService.Settings` to pin
  the previous default.
  (PR [#5246](https://github.com/pipecat-ai/pipecat/pull/5246))

- Updated the default model for `FishAudioTTSService` from `s2-pro` to
  `s2.1-pro`. Set `model` in `FishAudioTTSService.Settings` to pin the previous
  default.
  (PR [#5246](https://github.com/pipecat-ai/pipecat/pull/5246))

- Updated the default model for `LmntTTSService` from `aurora` to `blizzard`.
  Set `model` in `LmntTTSService.Settings` to pin the previous default.
  (PR [#5246](https://github.com/pipecat-ai/pipecat/pull/5246))

- Updated the default model for `AsyncAITTSService` and `AsyncAIHttpTTSService`
  from `async_flash_v1.0` to `async_flash_v1.5`. Set `model` in
  `AsyncAITTSService.Settings` or `AsyncAIHttpTTSService.Settings` to pin the
  previous default.
  (PR [#5246](https://github.com/pipecat-ai/pipecat/pull/5246))

- `SpeechTimeoutUserTurnStopStrategy` no longer overrides the deprecated
  `reset()` hook. Turn detection through `UserTurnController` is unaffected;
  code that calls `.reset()` directly on this strategy now reaches the
  inherited no-op instead — call `handle_user_turn_started()` (turn start) or
  `handle_user_turn_stopped()` (turn stop) instead.
  (PR [#5252](https://github.com/pipecat-ai/pipecat/pull/5252))

- Changed the default model for `GrokRealtimeLLMService` to
  `grok-voice-latest`, xAI's recommended Voice Agent alias. Pin a versioned
  model explicitly (e.g.
  `settings=GrokRealtimeLLMService.Settings(model="grok-voice-think-fast-1.0")`)
  for stability.
  (PR [#5255](https://github.com/pipecat-ai/pipecat/pull/5255))

- AWS Nova Sonic's `AudioConfig` now requires an `int` for each of its
  sample-rate, sample-size, and channel-count fields, rejecting an explicit
  `None` at construction. Every field already defaults to a real value, and a
  `None` that reached session continuation — which sizes its audio buffer from
  them — raised a `TypeError` there instead.
  (PR [#5273](https://github.com/pipecat-ai/pipecat/pull/5273))

- ⚠️ `LLMSetToolsFrame.tools` no longer lists a bare list of provider-specific
  tool dicts among the forms it accepts. That form last worked through
  `OpenAILLMContext`, which stored it verbatim, and stopped when that
  deprecated context was removed in 1.0.0. Provider-native tools travel in a
  `ToolsSchema`'s `custom_tools`, keyed by adapter type — a form the frame
  already carries end to end, into a realtime service's session update
  included. Nothing changes at runtime: frames are dataclasses and don't
  validate.
  (PR [#5273](https://github.com/pipecat-ai/pipecat/pull/5273))

- The async function-calling examples no longer set
  `enable_async_tool_cancellation=True`, so they demonstrate async tools on
  their own. Cancellation asks the model to judge whether a pending result is
  still wanted, and a model that judges too readily cancels a result the user
  was waiting on and never mentions it — which is worth knowing before turning
  it on, and is now noted on the parameter itself.
  (PR [#5278](https://github.com/pipecat-ai/pipecat/pull/5278))

- ⚠️ Removed Arcana model support from Rime TTS services before Rime's cloud
  cutoff on August 15, 2026 at 12:00 UTC. Set `model="coda"` when you upgrade.
  Rime examples use Luna for cross-model voice continuity.
  (PR [#5279](https://github.com/pipecat-ai/pipecat/pull/5279))

- ⚠️ Changed `SarvamLLMService` default `base_url` from
  `https://api.sarvam.ai/v1` to be resolved automatically from the selected
  model: `https://api.sarvam.ai/v1` for `sarvam-105b-conversations`,
  `https://api.sarvam.ai/v2` for all other models. Existing users who relied on
  the `/v1` default with `sarvam-105b` must pass
  `base_url="https://api.sarvam.ai/v1"` explicitly or update to `/v2`.
  (PR [#5288](https://github.com/pipecat-ai/pipecat/pull/5288))

- `FunctionCallCancelFrame` carries a `run_llm` field, defaulting to False.
  `LLMService` sets it only when a call is cancelled by its own timeout — an
  interruption must not trigger inference, and a cancellation the LLM requested
  already runs inference through the result of the tool that requested it.
  `LLMAssistantAggregator` pushes the context upstream when the flag is set,
  holding off while sibling calls from the same LLM response are still in
  flight so the group still triggers inference exactly once.
  (PR [#5291](https://github.com/pipecat-ai/pipecat/pull/5291))

- ⚠️ A function call that exceeds `function_call_timeout_secs` (or a per-tool
  `timeout_secs`) is now cancelled rather than left to run: its handler is
  thrown an `asyncio.CancelledError` so it can clean up, and the call settles
  through the path interruptions and LLM-requested cancellation already use — a
  `FunctionCallCancelFrame` and the `on_function_calls_cancelled` event — then
  runs inference so the bot reports that the call didn't complete. Previously
  the deadline reported an empty result while the handler kept running, so its
  side effects still landed and its real result was discarded. The deadline
  covers the handler's own execution; work it spawns into a task of its own is
  not cancelled with it.
  (PR [#5291](https://github.com/pipecat-ai/pipecat/pull/5291))

- ⚠️ `SonioxTTSService` now defaults to Soniox's `tts-rt-v2` model, with
  `Bryce` as the default voice. `tts-rt-v2` speaks the same WebSocket API as
  `tts-rt-v1` but offers a different roster of voices, so a `voice` set
  explicitly must be one `tts-rt-v2` offers. Soniox removes `tts-rt-v1` on
  August 31, 2026, after which requests naming it route to `tts-rt-v2`
  regardless.
  (PR [#5295](https://github.com/pipecat-ai/pipecat/pull/5295))

- `DeepgramFluxTTSService` now cancels the active turn with Flux's `Interrupt`
  message instead of reconnecting the websocket, so the cross-turn acoustic
  state that keeps a voice consistent survives a barge-in.

  Because an interruption no longer closes the connection, `on_connected` and
  `on_disconnected` stop firing on every barge-in.
  (PR [#5296](https://github.com/pipecat-ai/pipecat/pull/5296))

- Changed how a `function_call` expectation matches arguments in
  `pipecat.evals`. A turn expecting `args:` now passes if any call of that name
  matches them, where before it checked only the first call sharing the name
  and failed there. An LLM that gets a call wrong and immediately repeats it
  correctly now satisfies the turn, and when nothing matches, the failure names
  the arguments that did arrive.
  (PR [#5314](https://github.com/pipecat-ai/pipecat/pull/5314))

- A processor holds every frame it receives until its `StartFrame` arrives. A
  service that connects during setup can push frames before the pipeline
  starts; those frames now wait and are delivered after the `StartFrame`, in
  arrival order, so a processor never acts on a frame before it has started.
  (PR [#5316](https://github.com/pipecat-ai/pipecat/pull/5316))

- `StartupTimingObserver` measures a startup that now happens mostly before the
  `StartFrame`, so its report covers setting up as well as starting.

    - `ProcessorStartupTiming.duration_secs` is what a processor cost to get
      ready, its `setup()` and `start()` together, so it keeps reporting the same
      magnitude now that connecting has moved into `setup()`. The new
      `setup_duration_secs` breaks out the connecting part.
    - `StartupTimingReport.total_duration_secs` is the span from the pipeline
      starting to set up until it had started, rather than the sum of what each
      processor cost. Processors are set up concurrently, so a sum would report a
      pipeline as slower the more of its work overlapped.
    - `TransportTimingReport.bot_connected_secs` and `client_connected_secs`
      run from the pipeline starting to set up, so they measure the real time to a
      connected bot. A transport that connected before the `StartFrame` was pushed
      previously went unreported.
  (PR [#5316](https://github.com/pipecat-ai/pipecat/pull/5316))

- The pipeline clock now runs from the moment the pipeline starts setting up
  rather than from the `StartFrame`, so frames pushed while processors connect
  are no longer timestamped zero. Presentation timestamps therefore start at
  roughly what setting up cost; everything comparing them does so relatively,
  so pacing and playback are unaffected.
  (PR [#5316](https://github.com/pipecat-ai/pipecat/pull/5316))

- Pipeline configuration reaches processors through `FrameProcessorSetup` in
  `setup()` rather than through `StartFrame`. `setup.audio_in_sample_rate`,
  `setup.audio_out_sample_rate`, `setup.enable_metrics`,
  `setup.enable_tracing`, `setup.enable_usage_metrics`,
  `setup.report_only_initial_ttfb` and `setup.tracing_context` are available
  from `setup()` onwards, which is what lets a custom processor connect or
  resolve sample rates there.
  (PR [#5316](https://github.com/pipecat-ai/pipecat/pull/5316))

- `GroqLLMService` now defaults to `openai/gpt-oss-120b`. The previous default,
  `llama-3.3-70b-versatile`, is being retired by Groq. Pass
  `settings=GroqLLMService.Settings(model=...)` to choose a different model.
  (PR [#5338](https://github.com/pipecat-ai/pipecat/pull/5338))

- ⚠️ `UltravoxRealtimeLLMService` and `VonageVideoConnectorTransport` no longer
  cancel the pipeline when their connection fails. They now report the failure
  as one that leaves the service unusable, so the pipeline follows the
  `processor_unusable_policy` its `PipelineWorker` was given — by default it
  keeps running and the application decides what to do. Pass
  `processor_unusable_policy=ProcessorUnusablePolicy.CANCEL` to keep the
  previous behavior.
  (PR [#5348](https://github.com/pipecat-ai/pipecat/pull/5348))

- ⚠️ `GoogleVertexLLMService` now defaults to `gemini-3.6-flash`, and its
  default `location` changed from `us-east4` to `global`, because Vertex serves
  the Gemini 3 series only from the global endpoint. Pass
  `settings=GoogleVertexLLMService.Settings(model="gemini-2.5-flash")` and
  `location="us-east4"` to keep the previous configuration.
  (PR [#5356](https://github.com/pipecat-ai/pipecat/pull/5356))

- `DeepgramFluxSTTBase` moved to `pipecat.services.deepgram.flux.stt_base`.
  (PR [#5360](https://github.com/pipecat-ai/pipecat/pull/5360))

- OpenAI Realtime sessions now use `gpt-realtime-2.1` as the default model.
  (PR [#5362](https://github.com/pipecat-ai/pipecat/pull/5362))

- `AzureLLMService` now routes endpoints outside the v1 API surface through
  `2025-04-01-preview`, the last dated version Azure issued, so recent Azure
  features are available without naming a version.
  (PR [#5363](https://github.com/pipecat-ai/pipecat/pull/5363))

- The `cli` extra now requires Pipecat Context Hub 0.5.3 or newer, so a plain
  `pipecat-ai[cli]` install can run `pipecat context-hub refresh
  --framework-version latest` — the refresh the agent guides written by
  `pipecat init` prescribe. It pins the index to the newest released
  `pipecat-ai` tag instead of `main`, and re-resolves on every run, so a later
  incremental refresh picks up a new release without `--force`.
  (PR [#5367](https://github.com/pipecat-ai/pipecat/pull/5367))

- Updated the MoQ transport to `moq-rs` 0.4. Broadcasts are now created on an
  origin rather than constructed standalone, the track subscriptions
  (`subscribe_catalog`, `subscribe_audio`, `subscribe_json_stream`) are
  awaited, and `MoqError` is `Error`. The publish broadcast and transcript
  track are still created synchronously in `__init__`, so the bot loses no
  startup audio.

  Fixed the MoQ transport reporting a normal peer hangup as an error. A peer
  that vanishes mid-call drops its producer without finishing, which moq-rs
   raises with the reason as the message tail rather than as a reset code, so
  the hangup classifier missed it and the disconnect surfaced through
  `on_error` with a traceback.
  (PR [#5378](https://github.com/pipecat-ai/pipecat/pull/5378))

- `SarvamSTTService` now uses `saaras:v4` as its default model instead of
  `saaras:v3`. Applications that relied on the previous default should set
  `settings=SarvamSTTService.Settings(model="saaras:v3")` explicitly.
  (PR [#5382](https://github.com/pipecat-ai/pipecat/pull/5382))

- A TTS context that completes without producing any audio now resumes frame
  processing as soon as that is known, instead of leaving it paused until the
  pause watchdog fires a few seconds later. The non-fatal `ErrorFrame` the
  watchdog reports no longer accompanies these silent turns.
  (PR [#5393](https://github.com/pipecat-ai/pipecat/pull/5393))

- `TTSService` with `pause_frame_processing=True` now pauses only while there
  is audio to wait for: the bot speaking, or an audio context still open that
  may yet produce audio. Previously a turn that produced no audio could stall
  the pipeline for a few seconds until a watchdog force-resumed it and reported
  a non-fatal error.
  (PR [#5394](https://github.com/pipecat-ai/pipecat/pull/5394))

- The example bots set `processor_unusable_policy=ProcessorUnusablePolicy.END`,
  so an example ends once one of its processors can no longer do its job — a
  rejected API key or an unknown model, say — instead of running on with a
  service that will keep failing.
  (PR [#5397](https://github.com/pipecat-ai/pipecat/pull/5397))

- Changed `PipelineWorker.end()` and `PipelineWorker.activate_worker()` to wait
  for in-flight frames before they go through, so a closing line is heard
  rather than cut off and a worker handing over stops talking before the one
  taking over starts. Previously only `LLMWorker` arranged this. A pipeline
  that never started, or one that has already finished, is left alone.
  Cancelling still takes effect immediately.
  (PR [#5399](https://github.com/pipecat-ai/pipecat/pull/5399))

- Changed `WorkerRunner` to send each worker one shutdown message instead of
  two. `cancel()` now signals shutdown and the messages go out as the runner
  exits, carrying the reason the caller gave rather than a generic one, and
  addressed only to workers that have not already finished.
  (PR [#5399](https://github.com/pipecat-ai/pipecat/pull/5399))

- Changed `BaseWorker(active=...)` so that it governs whether a worker accepts
  bus messages at all. It previously gated only the frames a bridged worker
  received, leaving job requests, UI events and every other kind of bus traffic
  to arrive whatever the worker's state. An inactive worker is now handed only
  activation, deactivation, end or cancel messages. Nothing else reaches it, so
  no `on_bus_message` override or `on_bus_message` event handler runs for it
  either, which includes a `BaseUIWorker` no longer honouring the client's
  `__cancel_job_group` while inactive. `@worker_ready` handlers are unaffected,
  since they fire from the `WorkerRegistry` rather than over the bus.
  (PR [#5399](https://github.com/pipecat-ai/pipecat/pull/5399))

- Changed `PipelineWorker.flush_pipeline()` to wait for as long as the pipeline
  keeps working. Its `timeout` now counts seconds without progress rather than
  seconds in total, so a long turn keeps the wait alive while a stuck pipeline
  still gives up promptly. Progress is a frame reaching the sink, or a report
  from the pipeline answering the probe when it crossed into another worker.
  Heartbeats are not counted. A caller that gives up now says what it did:
  settled a function call before its output was delivered, or handed over
  without draining.
  (PR [#5399](https://github.com/pipecat-ai/pipecat/pull/5399))

- `KrispVivaSDKManager` now keeps the Krisp VIVA SDK initialized for the life
  of the process: `release()` no longer calls `krisp_audio.globalDestroy()`,
  and `is_initialized()` stays `True` after the last reference is released.
  Native sessions are still released per component, so per-call memory is
  unchanged. One consequence is that `api_key` is read only by the call that
  initializes the SDK, so a process serving sessions under different Krisp
  licenses uses the first one for all of them.
  (PR [#5411](https://github.com/pipecat-ai/pipecat/pull/5411))

- The `moonshine` extra now requires `moonshine-voice>=0.1.5`, up from
  `>=0.0.62`. Existing installs need `uv sync` (or `pip install -U
  "pipecat-ai[moonshine]"`) to pick the new version up.
  (PR [#5422](https://github.com/pipecat-ai/pipecat/pull/5422))

- Updated the `runner` extra to require `pipecat-ai-prebuilt>=1.0.6`,
  refreshing the prebuilt client UI served by the development runner with
  `@pipecat-ai/client-react` 1.8.2, `@pipecat-ai/moq-transport` 0.1.1, and
  `@pipecat-ai/voice-ui-kit` 0.13.1.
  (PR [#5427](https://github.com/pipecat-ai/pipecat/pull/5427))

- `AnthropicLLMService.ThinkingConfig` now covers Anthropic's current thinking
  API: `type="adaptive"`, the mode Claude 4.7 and later models require, and
  `display`, which asks for summarized thinking text on models that omit it by
  default.
  (PR [#5429](https://github.com/pipecat-ai/pipecat/pull/5429))

- A session now tears its controllers and its input audio filter down once,
  instead of once from the `EndFrame` or `CancelFrame` handler and again from
  `cleanup()`.
  (PR [#5434](https://github.com/pipecat-ai/pipecat/pull/5434))

- `VADController` and `UserTurnController` gained a `start()`, called by their
  owner, and they and `UserIdleController` gained a `stop()`.
  `BaseAudioFilter.start()` is now called from the input transport's `setup()`
  rather than on `StartFrame`.
  (PR [#5434](https://github.com/pipecat-ai/pipecat/pull/5434))

- ⚠️ Changed the user turn completion markers to a fill gradient: `●` marks a
  complete turn (previously `✓`), `◐` a turn cut off mid-thought (previously
  `○`), and `○` a user who needs more time (previously `◐`). The two incomplete
  markers have swapped meaning, so a custom
  `UserTurnCompletionConfig.instructions` string or a model fine-tuned on the
  old markers now maps short and long waits the wrong way round; set
  `complete_marker`, `incomplete_short_marker` and `incomplete_long_marker` on
  `UserTurnCompletionConfig` to keep the previous characters. Every marker is
  now a single token in every major tokenizer, and since the complete marker is
  generated before any speakable text, this removes up to two decode steps from
  the bot's first spoken word.
  (PR [#5437](https://github.com/pipecat-ai/pipecat/pull/5437))

- Changed `PipelineWorker.flush_pipeline()` to also wait for work the pipeline
  starts by pushing upstream, such as the LLM run a function call result
  triggers. The probe used to turn around at the source and settle there,
  returning before that response had been generated, let alone rendered; it now
  travels down, up, and down again, settling on the second arrival at the sink.
  (PR [#5438](https://github.com/pipecat-ai/pipecat/pull/5438))

- Changed `PipelineWorker.activate_worker()` to drain the pipeline only when
  `deactivate_self` is set. A worker that stays active is handing nothing over,
  so there is nothing in flight to wait for, and waiting meant the first
  activation of a session blocked on the very worker it was about to wake.
  (PR [#5438](https://github.com/pipecat-ai/pipecat/pull/5438))

- `pipecat init` now scaffolds Cartesia TTS with a voice recommended for
  `sonic-3.5`, the service's default model. Examples use the same voice.
  (PR [#5441](https://github.com/pipecat-ai/pipecat/pull/5441))

- `GradiumSTTService` now defaults `language` to `Language.EN` instead of
  leaving it unset. Grounding the model to a language improves transcription
  accuracy. Set `settings=GradiumSTTService.Settings(language="any")` to have
  Gradium detect the language instead.
  (PR [#5444](https://github.com/pipecat-ai/pipecat/pull/5444))

- `AnthropicLLMService` now disables thinking by default on Sonnet 5 and later,
  where adaptive thinking is otherwise on and the model decides per request
  whether to think, to keep latency low for real-time voice — mirroring how the
  Gemini service disables thinking by default on Flash models. Opus and Fable
  are left at Anthropic's default. Set `Settings.thinking` to configure
  thinking explicitly.
  (PR [#5446](https://github.com/pipecat-ai/pipecat/pull/5446))

- `CerebrasLLMService` now sends "developer"-role messages unchanged instead of
    converting them to "user" messages. Cerebras maps the role to its developer
    instruction layer, which sits above user instructions in the prompt
  hierarchy.
  (PR [#5448](https://github.com/pipecat-ai/pipecat/pull/5448))

- `MoondreamService` now defaults `revision` to `2025-06-21` instead of
  `2025-01-09`, picking up the newer Moondream build. Pass
  `revision="2025-01-09"` to stay on the previous one.
  (PR [#5458](https://github.com/pipecat-ai/pipecat/pull/5458))

### Deprecated

- Deprecated `MCPClient` methods `register_tools()`, `register_tools_schema()`,
  and `get_tools_schema()`. Use `MCPClient.tools()` instead.
  (PR [#4939](https://github.com/pipecat-ai/pipecat/pull/4939))

- Deprecated the `enable_user_speaking_frames` constructor parameter on
  `BaseUserTurnStartStrategy` and `BaseUserTurnStopStrategy`, which will be
  removed in 2.0.0. Whether a turn is announced is a per-turn decision rather
  than a per-strategy setting: pass `enable_user_speaking_frames` to
  `trigger_user_turn_started()` / `trigger_user_turn_stopped()` where the
  strategy decides the turn. Passing it to a constructor still applies and now
  emits a `DeprecationWarning`.

  `ExternalUserTurnStartStrategy` and `ExternalUserTurnStopStrategy` suppress
  emission on their own whenever the turn was already announced elsewhere — by
  a shared `UserTurnProcessor`, or by a service that emits turn frames rather
  than proposing them — so a pipeline built on those strategies doesn't need to
  set the flag anywhere.
  (PR [#5156](https://github.com/pipecat-ai/pipecat/pull/5156))

- Deprecated `UIWorker.ui_job_group()`, `UIWorker.start_ui_job_group()`, and
  `UIJobGroupContext` (all removed in 2.0.0): use `job_group(...)` /
  `request_job_group(...)` / `JobGroupContext` instead, since every group a
  `BaseUIWorker` dispatches is client-visible. The deprecated wrappers keep
  their historical signatures and behavior in the meantime.
  (PR [#5221](https://github.com/pipecat-ai/pipecat/pull/5221))

- Deprecated passing `name`, `payload`, `timeout`, and `cancel_on_error`
  directly to `BaseWorker.job()`, `job_group()`, `request_job()`,
  `request_job_group()`, and `create_job_group_and_request_job()` (removed in
  2.0.0). Pass `params=JobParams(...)` or `params=JobGroupParams(...)` instead.
  The individual arguments keep working in the meantime, and passing both
  raises `TypeError`.
  (PR [#5221](https://github.com/pipecat-ai/pipecat/pull/5221))

- Deprecated the `cartesia_version` parameter of `CartesiaTTSService` and
  `CartesiaHttpTTSService`. Both services send the `Cartesia-Version` header
  they are written against, since their request payloads and response handling
  are tied to that version. Passing `cartesia_version` warns and still
  overrides the header until it is removed in 2.0.0.
  (PR [#5231](https://github.com/pipecat-ai/pipecat/pull/5231))

- Deprecated `enable_async_tool_cancellation` on LLM services; it will be
  removed in 2.0.0. Set `cancellable_by_llm=True` on the tools that should be
  cancellable instead. It still works meanwhile, treating every async tool as
  cancellable — which is worth moving off, because a model that wrongly decides
  a pending result is unwanted destroys work the user asked for, and a tool
  that never opted in can't have that happen to it.

  The flag's shape has changed with it: where it used to advertise a single
  generic cancel tool, it now advertises a `cancel_<name>` for every async
  tool, so the tool set a model sees grows with the number of async tools
  registered.

      ```python
      # Before
      llm = OpenAILLMService(api_key=..., enable_async_tool_cancellation=True)


      @tool_options(cancel_on_interruption=False)
      async def write_report(params: FunctionCallParams, topic: str): ...


      # After
      llm = OpenAILLMService(api_key=...)


      @tool_options(cancel_on_interruption=False, cancellable_by_llm=True)
      async def write_report(params: FunctionCallParams, topic: str): ...
      ```
  (PR [#5304](https://github.com/pipecat-ai/pipecat/pull/5304))

- Deprecated `StartFrame.audio_in_sample_rate`,
  `StartFrame.audio_out_sample_rate`, `StartFrame.enable_metrics`,
  `StartFrame.enable_tracing`, `StartFrame.enable_usage_metrics`,
  `StartFrame.report_only_initial_ttfb` and `StartFrame.tracing_context`, which
  will be removed in 2.0.0. Read the same values from `FrameProcessorSetup` in
  `setup()` instead. The fields still carry the pipeline's configuration, so a
  processor that reads one keeps working and emits a `DeprecationWarning`, once
  per call site.
  (PR [#5316](https://github.com/pipecat-ai/pipecat/pull/5316))

- `AudioBufferProcessor`'s `on_user_turn_audio_data` and
  `on_bot_turn_audio_data` are deprecated and will be removed in 2.0.0. They
  report a run of speech at a time, so one turn produces several and none
  carries a turn number. Use `on_user_turn_audio` and `on_bot_turn_audio`
  instead.
  (PR [#5329](https://github.com/pipecat-ai/pipecat/pull/5329))

- Deprecated "fatal" errors. Concretely, deprecated 3 things:
  `ErrorFrame.fatal`, the `fatal` argument of `FrameProcessor.push_error()`,
  and `FatalErrorFrame`, all of which will be removed in 2.0.0. A fatal error
  would cancel the pipeline outright; that's now an application decision.
  Passing `fatal=True` still cancels the pipeline, but now also emits a
  `DeprecationWarning`. There are two alternatives to fatal errors, depending
  on what your error means:

    - The error leaves its originating processor unable to do any more work:
      report it with `push_error(..., force_treat_as_permanent=True)`. That marks
      the processor unusable, and `PipelineWorker` applies its
      `processor_unusable_policy` to specify how to handle the resulting error.

    - The error isn't about any processor's state, but the pipeline should stop
      anyway: push a regular `ErrorFrame` (without `fatal`) and follow it with an
      `EndWorkerFrame`, which ends the pipeline after queued frames drain. Use
      `CancelWorkerFrame` instead to abandon the queued frames, as `fatal=True`
      did.
  (PR [#5348](https://github.com/pipecat-ai/pipecat/pull/5348))

- Deprecated the `api_version` constructor parameter on `AzureLLMService`,
  which will be removed in 2.0.0. Point `endpoint` at Azure's v1 API surface
  instead, by ending it in `/openai/v1`. Azure issued no dated version after
  `2025-04-01-preview`, and new features reach only the v1 surface. Endpoints
  outside that surface still route through `2025-04-01-preview`; passing
  `api_version` explicitly still applies and now emits a `DeprecationWarning`.
  (PR [#5363](https://github.com/pipecat-ai/pipecat/pull/5363))

- Deprecated the `pause_watchdog_timeout_s` parameter of `TTSService`, which
  will be removed in 2.0.0. Passing it warns and does nothing: a pause is now
  only taken while audio is playing or still on its way, so it is always lifted
  by the `BotStoppedSpeakingFrame` that follows playback or by the audio
  context completing in silence — no timer is needed to break it.
  (PR [#5394](https://github.com/pipecat-ai/pipecat/pull/5394))

- Deprecated the `target_task` parameter of `BusBridgeProcessor`. Use
  `target_worker` instead; a "task" is an asyncio task and the thing being
  named here is a worker. Passing `target_task` still works and emits a
  `DeprecationWarning`. It will be removed in 2.0.0.
  (PR [#5438](https://github.com/pipecat-ai/pipecat/pull/5438))

- Deprecated the `messages` and `result_callback` parameters of
  `LLMWorker.end()` and `LLMWorker.activate_worker()`. Deliver the function
  call result from the tool handler instead, with `await
  params.result_callback(result)`, and the output it triggers is delivered
  before the worker ends or hands over. Passing either parameter still works
  and emits a `DeprecationWarning`. They will be removed in 2.0.0.
  (PR [#5438](https://github.com/pipecat-ai/pipecat/pull/5438))

### Removed

- Removed `AICVADAnalyzer`, `AICFilter.create_vad_analyzer()`, and
  `AICFilter.get_vad_context()`. `aic-sdk` 3.0 removed the energy-based VAD all
  three relied on. The first two were deprecated since 1.4.0;
  `get_vad_context()` was not, so calls to it need replacing with
  `AICQuailVADAnalyzer`, which runs a dedicated VAD model.
  (PR [#5239](https://github.com/pipecat-ai/pipecat/pull/5239))

- Removed the sunset `saarika:v2.5` and `saaras:v2.5` models from
  `SarvamSTTService`, leaving `saaras:v3` and `saaras:v4` as the supported
  models; applications pinned to either should move to `saaras:v4`. The service
  now always connects to the transcription endpoint, since
  `speech_to_text_translate_streaming` only served `saaras:v2.5` — translation
  is still available on the remaining models through `mode="translate"`.
  (PR [#5383](https://github.com/pipecat-ai/pipecat/pull/5383))

- ⚠️ Removed the `prompt` setting and the `set_prompt()` method from
  `SarvamSTTService`. Both were only ever honored by `saaras:v2.5`, which
  Sarvam is sunsetting, so there is no replacement — code passing
  `SarvamSTTService.Settings(prompt=...)` should drop the argument.
  (PR [#5383](https://github.com/pipecat-ai/pipecat/pull/5383))

### Fixed

- Fixed deadlock caused by `FrameProcessorResumeFrame` waiting in the process
  queue by changing it to a `SystemFrame`.
  (PR [#3448](https://github.com/pipecat-ai/pipecat/pull/3448))

- Made `GoogleLLMService`, `GoogleVertexLLMService`, and `GeminiLiveLLMService`
  more resilient to a tool's JSON schema using a construct Gemini doesn't
  accept. Gemini supports only a limited subset of JSON Schema, and a single
  tool with an unsupported construct would fail the entire request.
  `GeminiLLMAdapter` now tries to convert the tool schemas into the supported
  subset before the request, logging each change it makes:
    - Vendor extensions (`x-` prefixed keys, such as the `x-mcp-header`
      GitHub's MCP server attaches to most of its tool properties) are dropped,
      joining the `additionalProperties` already stripped.
    - A union `type`, such as `["string", "number"]`, becomes the equivalent
      `anyOf`.
    - An `enum` whose members aren't strings is dropped, losing its constraint.
  (PR [#4939](https://github.com/pipecat-ai/pipecat/pull/4939))

- Fixed `MCPClient` methods `start()` and `tools()` hanging indefinitely when a
  server refused the connection. A failing transport cancels the connecting
  task from inside its own task group, and that cancellation went uncaught,
  leaving the connection result unsettled. The underlying error is now raised
  to the caller.
  (PR [#4939](https://github.com/pipecat-ai/pipecat/pull/4939))

- Fixed `OpenAIResponsesHttpLLMService` producing a silent, empty turn when the
  Responses API reported a `response.failed`, `response.incomplete`, or `error`
  event mid-stream. These events arrive on an otherwise healthy stream, so
  nothing raised and no `ErrorFrame` was pushed, leaving
  `ServiceSwitcherStrategy` unable to fail over and the failure absent from
  logs. They now push an `ErrorFrame`, matching the WebSocket variant.
  (PR [#5141](https://github.com/pipecat-ai/pipecat/pull/5141))

- Fixed a reconnect that could be deferred forever on an STT service with
  built-in turn detection. `STTService` defers a reconnect requested while the
  user is speaking and re-enables it on `UserStoppedSpeakingFrame`, but a
  service that emitted that frame itself never received one — a broadcast
  doesn't reach its own emitter — so with a VAD analyzer in the pipeline the
  deferred reconnect never fired. These services now propose turn boundaries
  and the user aggregator emits the turn frames, which do reach the service.
  (PR [#5156](https://github.com/pipecat-ai/pipecat/pull/5156))

- Fixed the MoQ transport reporting an ordinary hangup as a transport failure.
  A peer disconnecting resets every in-flight track subscription, which
  surfaces as a per-track error carrying a numeric remote code — distinct from
  the session-level WebTransport close the transport already recognised. A
  browser leaving mid-call drops its microphone producer without finishing it,
  so the bot's audio subscriber saw a `Dropped` reset and logged an `ERROR`
  plus a traceback and invoked `on_error`, for what is just the end of the
  call. Peer-gone reset codes are now treated as a normal close, like the
  session-level one.
  - Constrained the MoQ extra to `moq-rs~=0.3.2`. The previous `<1.0.0` bound
  bought nothing against a hand-versioned pre-1.0 library: 0.4.0 renamed
  `MoqError` to `Error`, replaced `OriginProducer.publish()` with
  `create_broadcast()`, and made the `subscribe_*` helpers async, so a fresh
  install resolved to a release the transport can't run on.
  - Fixed the MoQ transport dropping its producers instead of finishing them on
  disconnect. Finishing the audio track flushes samples still inside the
  encoder, and finishing the broadcast unannounces it — dropped, it gets
  lingered instead, so the relay kept advertising a dead bot after every call.
  (PR [#5158](https://github.com/pipecat-ai/pipecat/pull/5158))

- Fixed `WhisperSTTService` silently transcribing in English when its model
  can't handle the configured language. The English-only models — every `.en`
  one, including the default `distil-medium.en` — accept any language and
  transcribe as English regardless, so `Settings(language=Language.ES)`
  produced fluent-looking English rather than an error. Constructing such a
  pairing now raises a `ValueError` naming the model and its supported
  languages; a mid-call switch via `STTUpdateSettingsFrame` reports a non-fatal
  `ErrorFrame` instead, leaving the pipeline running.

  ⚠️ Code that set a non-English `language` on an English-only model was
  getting English transcripts and now raises at construction. Use a
  multilingual model (e.g. `large-v3-turbo`) or drop the `language`.
  (PR [#5171](https://github.com/pipecat-ai/pipecat/pull/5171))

- Fixed `KokoroTTSService` failing to synthesize French and Mandarin.
  kokoro-onnx phonemizes through espeak-ng, which has no `zh` and no bare `fr`
  voice, so both raised `language "..." is not supported by the espeak backend`
  at synthesis time. Mandarin (including the `zh-CN`/`zh-HK`/`zh-TW` variants)
  now maps to `cmn` and French to `fr-fr`, with `fr-be`, `fr-ch` and `pt-br`
  mapped to the regional espeak-ng voices they have.
  (PR [#5171](https://github.com/pipecat-ai/pipecat/pull/5171))

- Fixed `MoonshineSTTService` failing to construct for any non-English
  language. Moonshine publishes its streaming architectures for English only
  and most other languages ship a single model, so the default
  `small-streaming` architecture didn't exist for, say, Spanish. An
  architecture unavailable for the configured language now falls back to the
  best model published for it.
  (PR [#5182](https://github.com/pipecat-ai/pipecat/pull/5182))

- Fixed services surviving pipeline teardown and reconnecting as orphans.
  `TaskManager.cancel_task()` absorbed every `CancelledError` raised while
  awaiting the task it had cancelled, including the calling task's own
  cancellation. Because asyncio delivers a cancellation only once, a service
  tearing down from a `finally` block —
  `DeepgramSTTService._connection_handler` cancelling its keepalive, for
  example — never learned it had been cancelled, and as a reconnect loop went
  on reconnecting unsupervised. `cancel_task()` now re-raises a cancellation
  delivered to the caller while it waits, and still absorbs the cancelled
  task's own.
  (PR [#5186](https://github.com/pipecat-ai/pipecat/pull/5186))

- Fixed `expand_units` reading a quantity of one with a plural unit, so "Only
  1km left" now becomes "Only 1 kilometer left" instead of "Only 1 kilometers
  left". A decimal such as "1.0km" keeps the plural.
  (PR [#5205](https://github.com/pipecat-ai/pipecat/pull/5205))

- Fixed `ElevenLabsRealtimeSTTService` pushing two final `TranscriptionFrame`s
  per utterance when `include_language_detection` was enabled without
  `include_timestamps`.
  (PR [#5208](https://github.com/pipecat-ai/pipecat/pull/5208))

- Fixed `expand_numbers` dropping a decimal's trailing zero, so "1.0" now reads
  as "one point zero" instead of the bare "one". This was most audible composed
  with `expand_units`, which keeps the plural for a decimal:
  `VoiceFormatter(expand_numbers=True)` turned "1.0km left" into "one
  kilometers left". Decimals without trailing zeros are unchanged.
  (PR [#5213](https://github.com/pipecat-ai/pipecat/pull/5213))

- Fixed `ExotelFrameSerializer` sending the stream identifier as `streamSid` on
  outbound `media` and `clear` events. Exotel's media stream protocol spells it
  `stream_sid`. Exotel treats the identifier as optional on messages from the
  bot, so existing integrations were unaffected.
  (PR [#5219](https://github.com/pipecat-ai/pipecat/pull/5219))

- Fixed an async function call's result going unreported when the conversation
  moved on while the call was still running. A tool registered with
  `cancel_on_interruption=False` keeps running after the LLM's turn ends, so by
  the time its result arrives the user has often changed the subject — and the
  LLM would answer the new topic without ever mentioning the result. The
  final-result message now instructs the model to finish responding to whatever
  the user is talking about and then deliver the result at the end of that
  response, stating a short result outright and naming a long one with an offer
  of the details.
  (PR [#5236](https://github.com/pipecat-ai/pipecat/pull/5236))

- Fixed `push_error_frame()` raising an unrelated `IndexError` in place of the
  error being reported, when that error carried an exception that was never
  raised and so had no traceback to read.
  (PR [#5242](https://github.com/pipecat-ai/pipecat/pull/5242))

- Fixed `DeepgramSTTService` dropping the speaker's first word or two when
  someone is already talking as a session starts. The connection is established
  in the background, so audio arriving before there was a connection to carry
  it was discarded. Frames now wait at the service until the connection can
  carry them, and are transcribed in full once it can.
  (PR [#5254](https://github.com/pipecat-ai/pipecat/pull/5254))

- Fixed `GrokRealtimeLLMService` dropping xAI Voice Agent server events that
  were not registered in the parser (notably `session.created` on every
  connect). The service now parses the full documented server event set, pushes
  interim user captions from
  `conversation.item.input_audio_transcription.updated`, and handles
  text-modality deltas from `response.text.delta` /
  `response.output_text.delta`.
  - Fixed `GrokRealtimeLLMService` silently dropping user audio while
  conversation seeding was pending. Audio now flows after `session.updated`, so
  audio-only pipelines work without an explicit `LLMRunFrame` /
  `_create_response`.
  - Fixed interruptions under server VAD not cancelling the in-flight response
  on the wire. `InterruptionFrame` now always sends `response.cancel`; the
  input buffer is cleared only in manual turn mode so interrupting user speech
  is preserved.
  - Fixed `GrokRealtimeLLMService` interruptions only clearing local audio
  state. Interruptions now also send `conversation.item.truncate` so
  server-side conversation history matches what the user heard.
  (PR [#5255](https://github.com/pipecat-ai/pipecat/pull/5255))

- Fixed `CartesiaTTSService` and `SonioxTTSService` dropping an already-heard
  sentence prefix from the transcript when a voice/model/language (or, for
  Soniox, speed) settings change was applied mid-sentence. The re-mint of the
  turn context now finalizes the old context's pending sentence first — so
  word-timestamps arriving during the flushed playout still emit
  `AggregatedTextProgressFrame`s — mirroring the existing end-of-turn and
  `TTSSpeakFrame` close paths.
  (PR [#5257](https://github.com/pipecat-ai/pipecat/pull/5257))

- Fixed eval turns matching — and judges ruling on — output the bot produced
  for an earlier turn. Events queue up between turns and the matcher starts
  consuming as soon as a turn's input is sent, so whatever was already waiting
  was read first; a turn with `send_after` made the window seconds wide. A turn
  that sends input now drops the queued bot output first. Turns that send
  nothing are observation-only and exist to match exactly that pending output,
  so they keep it.
  (PR [#5260](https://github.com/pipecat-ai/pipecat/pull/5260))

- `GoogleLLMService` now closes a Gemini stream it stops consuming, so an
  interrupted or timed-out response releases its HTTP resources right away
  instead of waiting on garbage collection.
  (PR [#5262](https://github.com/pipecat-ai/pipecat/pull/5262))

- Fixed `PatternPairAggregator` and `SkipTagsAggregator` mishandling an LLM
  response that ends with an unclosed start tag. `PatternPairAggregator.flush()`
  no longer leaks REMOVE-pattern content to TTS: it cuts at the earliest
  truly-unmatched REMOVE/AGGREGATE start delimiter (keeping unclosed KEEP
  content verbatim) and trims a trailing partial start delimiter.
  `SkipTagsAggregator.flush()` in TOKEN mode now returns buffered text instead
  of silently dropping it.
  (PR [#5266](https://github.com/pipecat-ai/pipecat/pull/5266))

- Fixed `PatternPairAggregator` and `SkipTagsAggregator` in TOKEN mode
  mishandling a start delimiter split across `aggregate()` calls: a trailing
  partial start delimiter is now held back until the next chunk completes it
  instead of being flushed (and spoken) as plain text. Also fixed
  `SkipTagsAggregator` losing track of its tag-scan position after a TOKEN-mode
  yield, which made every tag after the first closed one go undetected.
  (PR [#5268](https://github.com/pipecat-ai/pipecat/pull/5268))

- Fixed `OpenAIResponsesHttpLLMService` running a function call with fabricated
    empty arguments when the stream ended in a terminal error
  (`response.failed`,
    `response.incomplete`, or `error`) before the call's arguments finished
    streaming. Calls whose arguments did finish streaming still run.
  (PR [#5270](https://github.com/pipecat-ai/pipecat/pull/5270))

- Fixed `GoogleTTSService` and `GoogleHttpTTSService` raising `TypeError` on a
  settings update that set `speaking_rate` to `None`. `None` is the field's
  default and the way to leave the rate to Google, but the range check these
  services run on an incoming rate handed it to `float()`. A `None` rate now
  skips the range check.
  (PR [#5273](https://github.com/pipecat-ai/pipecat/pull/5273))

- Fixed `InworldRealtimeLLMService` raising `ValueError` when its input or
  output audio format was PCMU or PCMA. On every start the service syncs the
  configured format's sample rate with the transport's, and the G.711 formats
  are fixed at 8000 Hz and declare no rate to write to. The sync now applies
  only to the PCM format, the one with a configurable rate.
  (PR [#5273](https://github.com/pipecat-ai/pipecat/pull/5273))

- Fixed `SpeechmaticsSTTService` raising `AttributeError` when constructed with
  an English locale it has no output-locale mapping for, such as
  `Language.EN_IN`. Such a locale is meant to log a warning and fall back to
  the base language code, but composing that warning was itself what raised.
  Construction now succeeds and the fallback is logged.
  (PR [#5273](https://github.com/pipecat-ai/pipecat/pull/5273))

- Fixed `GeminiTTSService` raising `AttributeError` on a settings update typed
  as the base `TTSSettings` rather than `GeminiTTSService.Settings`. The
  service reads `multi_speaker` and `prompt` off the delta to warn about
  settings its GenAI backend ignores, and those fields exist only on its own
  settings type. It now reads them only when the delta carries them, as the
  sibling Google TTS services already do.
  (PR [#5273](https://github.com/pipecat-ai/pipecat/pull/5273))

- Fixed `SimliVideoService` raising `AttributeError` when using
  `is_trinity_avatar=True` due to its calling a nonexistent
  method—`playImmediate`—on the Simli client. The intended method is called
  `sendImmediate`.
  (PR [#5273](https://github.com/pipecat-ai/pipecat/pull/5273))

- Bots with async tool cancellation enabled now emit the
  `cancel_async_tool_call` call rather than only acknowledging the cancellation
  out loud. The instructions given to the LLM state that the call is the only
  thing that stops the pending work, so a bot that says it will skip a result
  no longer has that result arrive moments later and contradict it.
  (PR [#5276](https://github.com/pipecat-ai/pipecat/pull/5276))

- `enable_async_tool_cancellation=True` now takes effect for bots that declare
  their tools through an `LLMContext`, which covers the direct-function and
  `FunctionSchema` handler patterns. Previously the built-in
  `cancel_async_tool_call` tool was never advertised to the LLM in that case —
  setup ran before those handlers were registered — so a bot could not cancel
  an async function call whose result the user no longer wanted, however
  clearly they asked for it. Setup no longer depends on a handler being
  registered before the pipeline starts.
  (PR [#5276](https://github.com/pipecat-ai/pipecat/pull/5276))

- Fixed an async function call being made a second time, with a fabricated
  result, while the first was still running. The message announcing the call to
  the model described the message its result would arrive in — the role, the
  fields, how many there might be — and a model told the shape of a message it
  should expect tries to produce one, through the only structured channel it
  has: another function call, carrying the protocol payload as its arguments.
  The announcement now says only that the task is running, that its result will
  be given to the model, and that it should neither call again nor answer from
  nothing. Most visible on `GoogleLLMService`, where the description named a
  developer-role message that the Gemini adapter rewrites to a user message, so
  the shape it described never arrived at all.
  (PR [#5277](https://github.com/pipecat-ai/pipecat/pull/5277))

- An async function call's result is now reported reliably when the
  conversation has moved on, and reported after the answer to whatever the user
  last asked rather than ahead of it. A bot registering a tool with
  `cancel_on_interruption=False` gets standing guidance in its system
  instruction — a result that has arrived is owed to the user, it belongs at
  the end of the reply that answers them, and it is said once. The per-result
  message carried the same policy, but it arrives buried in a context whose
  most recent turn is the user asking for something else, and a model weighing
  the two would answer and leave the result unsaid, or state it before the
  answer.
  (PR [#5278](https://github.com/pipecat-ai/pipecat/pull/5278))

- Fixed a function call whose handler raises never being settled. The exception
  was reported upstream as a non-fatal `ErrorFrame` and then nothing else
  happened, so the call stayed in progress forever:
  `has_function_calls_in_progress` never cleared, sibling calls from the same
  LLM response could no longer complete their group, and a
  `FunctionCallUserMuteStrategy` or `UserIdleController` counting the call
  never saw it finish. The call now settles with a result reporting that the
  function failed, so the LLM can tell the user; the exception stays on the
  `ErrorFrame` and out of the LLM context.
  (PR [#5291](https://github.com/pipecat-ai/pipecat/pull/5291))

- Fixed a cancelled async function call (registered with
  `cancel_on_interruption=False`) never being settled in the LLM context. It
  stayed in progress forever, so `has_function_calls_in_progress` never cleared
  and inference was suppressed for the rest of a parallel tool-call group.
  Cancelling one now settles it the way synchronous tool calls settle.
  (PR [#5291](https://github.com/pipecat-ai/pipecat/pull/5291))

- Fixed a late result from a function call handler being broadcast into the
  pipeline only for the aggregator to log a warning and drop it. `LLMService`
  now rejects results for a call already settled by a final result, a timeout,
  or a cancellation.
  (PR [#5291](https://github.com/pipecat-ai/pipecat/pull/5291))

- Fixed the sequential function call runner (`run_in_parallel=False`) shutting
  down when an in-flight call was cancelled, which left every later function
  call in the conversation unexecuted.
  (PR [#5291](https://github.com/pipecat-ai/pipecat/pull/5291))

- Fixed `LiveKitTransport` identifying participants by LiveKit's `sid` (a
  per-connection session id) everywhere it surfaces a `participant_id` — event
  handlers, `LiveKitInputTransportMessageFrame`, `get_participants()` — while
  `get_participant_metadata()`, `mute_participant()`, and
  `unmute_participant()` look the id up in `room.remote_participants`, which
  LiveKit keys by `identity` instead. The id `get_participants()`/events handed
  out could never be fed into those three lookup methods, so
  `get_participant_metadata()` silently returned `{}` and
  `mute_participant()`/`unmute_participant()` silently did nothing.
  `participant_id` is now consistently the participant's LiveKit identity
  throughout. Those three methods also referenced `is_speaking` and `tracks`,
  attributes the current `livekit` SDK no longer has (`track_publications`
  replaces `tracks`); `get_participant_metadata()` no longer includes
  `is_speaking`, and muting now unsubscribes from the participant's audio track
  via `track_publications`.
  (PR [#5297](https://github.com/pipecat-ai/pipecat/pull/5297))

- Fixed `LiveKitTransport` never delivering client messages (including RTVI's
  `client-ready` handshake) to the pipeline. Incoming data-channel messages
  were wrapped in an output-message frame and pushed downstream only, so
  `RTVIProcessor` never saw them — instead, the output transport picked the
  misrouted frame back up and echoed it straight back out to the room. Messages
  are now parsed and broadcast as `InputTransportMessageFrame` in both
  directions, matching Daily and SmallWebRTC, so RTVI-based bots using LiveKit
  now complete the client-ready/bot-ready handshake and receive client messages
  correctly. Non-JSON or non-object data on the channel is ignored rather than
  raising, and still fires `on_data_received` for backwards compatibility.
  (PR [#5297](https://github.com/pipecat-ai/pipecat/pull/5297))

- Fixed xAI STT to use the pipeline input sample rate when no explicit rate is
  configured.
  (PR [#5298](https://github.com/pipecat-ai/pipecat/pull/5298))

- Fixed AssemblyAI STT to use the pipeline input sample rate when no explicit
  rate is configured.
  (PR [#5298](https://github.com/pipecat-ai/pipecat/pull/5298))

- Fixed Krisp VIVA support against SDK 1.11.0 and newer, which switched its
  Python bindings from pybind11 to nanobind. `KrispVivaFilter`,
  `KrispVivaTurn`, and `KrispVivaIPUserTurnStartStrategy` detect the binding
  style at runtime, so both older and newer SDK builds work.
  `KrispVivaFilter`'s `noise_suppression_level` is now a float; an int is still
  accepted.
  (PR [#5302](https://github.com/pipecat-ai/pipecat/pull/5302))

- Fixed intermittent failures in tests written with `run_test()`. Frames were
  sent after a fixed 10ms delay, so a pipeline that took longer than that to
  start would drop them; `run_test()` now waits for the pipeline to be ready
  before sending. A new `start_timeout` argument (1 second by default) raises
  `TimeoutError` if the pipeline never starts.
  (PR [#5313](https://github.com/pipecat-ai/pipecat/pull/5313))

- A service that fails to connect is left unable to do its job, so a
  `ServiceSwitcher` moves off it before the pipeline starts and every frame
  reaches a service that connected. Setting up is not attempted again, so the
  failure is permanent whatever caused it: a connection timeout previously left
  the service usable and the switcher on it.
  (PR [#5316](https://github.com/pipecat-ai/pipecat/pull/5316))

- A processor that raises while setting up now pushes an `ErrorFrame` upstream,
  the same way a failure while handling a frame is reported, so application
  code learns its pipeline came up degraded. The error was previously only
  logged and the pipeline ran on regardless. Each failing processor reports its
  own error, so a pipeline where several fail reports all of them rather than
  only whichever raised first.
  (PR [#5316](https://github.com/pipecat-ai/pipecat/pull/5316))

- `DeepgramSTTService` stops reconnecting once three attempts in a row have
  failed to produce a connection that stays up, and reports itself unusable so
  a `ServiceSwitcher` moves off it. A handshake that hung before failing, or a
  connection that dropped after a while, previously reset the count and it
  retried for the life of the process.
  (PR [#5316](https://github.com/pipecat-ai/pipecat/pull/5316))

- A processor that raises while being cleaned up no longer costs the rest of
  the pipeline its teardown. Each failure is logged and every other processor
  is still released.
  (PR [#5316](https://github.com/pipecat-ai/pipecat/pull/5316))

- Fixed TTFB being measured inconsistently across LLM services, so the values
  were not comparable between them. TTFB now represents the time to the first
  byte of the model's streamed response for every LLM service.
  `AnthropicLLMService` and `AWSBedrockLLMService` stopped measuring as soon as
  the stream was created, before reading any event, so their TTFB reflected
  connection setup rather than the model's response; `GoogleLLMService` stopped
  on the first chunk, which can carry usage metadata and no model output.
  Reasoning is part of the response, so a thinking model's TTFB ends at its
  first reasoning token.

  TTFB values for these models may be increased.
  (PR [#5319](https://github.com/pipecat-ai/pipecat/pull/5319))

- The eval judge waits for a bot that is still working instead of failing the
  turn. A bot with async tools acknowledges a request and answers once its tool
  returns, and the acknowledgement was being judged as a wrong answer: whether
  a reply counted as an answer turned on how it read, so a fluent "The system
  is checking the current conditions for you right now." was taken for one. A
  bot that says it is checking, fetching, or will report back has not answered
  yet, whatever the length or polish of the sentence it says it in.
  (PR [#5328](https://github.com/pipecat-ai/pipecat/pull/5328))

- Fixed synthesis markup disappearing from a turn's spoken text when no word
  follows it. A tag closing a sentence — or sitting between the last word and
  its period — is never named by a word-timestamp event, so it was missing from
  `AggregatedTextProgressFrame.accumulated_text` and from the text the turn
  reported as spoken.
  (PR [#5331](https://github.com/pipecat-ai/pipecat/pull/5331))

- Fixed word-level TTS tracking stopping partway through a sentence containing
  synthesis markup, when the provider punctuates a tagged span differently from
  the source text.
  (PR [#5331](https://github.com/pipecat-ai/pipecat/pull/5331))

- Fixed a sentence losing word-level TTS tracking when the synthesis markup
  comes from the LLM itself, e.g. an LLM prompted to emit `<spell>1234</spell>`
  with `SkipTagsAggregator` keeping the tagged block intact. The whole sentence
  was treated as one untrackable unit: it reported no progress until it had
  finished speaking, and its words reached the conversation context only as a
  single block at the end. Now only the tagged span is committed whole, so
  every word around it gets its own `TTSTextFrame` and
  `AggregatedTextProgressFrame`. Applies to both `SENTENCE` and `TOKEN` text
  aggregation. Tags inserted by a text transform were never affected, since
  those reach the TTS without appearing in the user-facing text.
  (PR [#5331](https://github.com/pipecat-ai/pipecat/pull/5331))

- Fixed VAD analyzers, turn analyzers, local audio and Tk output transports,
  and the Daily and Vonage clients leaking a worker thread per session, plus
  one per output destination of every transport. The thread pools they run
  blocking work on are now shut down at cleanup.
  (PR [#5350](https://github.com/pipecat-ai/pipecat/pull/5350))

- Fixed `GoogleLLMService` being unusable with `gemini-3.7-flash`, which
  rejects the `minimal` thinking level Pipecat applies as a low-latency default
  (every request failed with `400 INVALID_ARGUMENT`). It now gets `low`, the
  lowest level it accepts.
  (PR [#5356](https://github.com/pipecat-ai/pipecat/pull/5356))

- Fixed word-level TTS tracking breaking when a TTS service normalizes
  typographic punctuation in its word-timestamp events, e.g. reporting `don't`
  for a `don’t` it was sent, or the reverse, and likewise for curly quotes and
  en/em dashes.
  (PR [#5357](https://github.com/pipecat-ai/pipecat/pull/5357))

- Pinless dial-in update failures now trigger the Daily transport's `on_error`
  event instead of only being logged.
  (PR [#5358](https://github.com/pipecat-ai/pipecat/pull/5358))

- The `DeprecationWarning` for reading `FrameProcessorSetup.tool_resources` is
  reported once per call site, and points at the line that performed the read.
  A caller that reflects over every field of every object it sees — a frame
  serializer, for example — previously repeated the warning without bound.
  (PR [#5365](https://github.com/pipecat-ai/pipecat/pull/5365))

- Fixed the Google LLM services ignoring the `seed` setting.
  `GoogleLLMService`, `GoogleVertexLLMService`, `GeminiLiveLLMService`, and
  `GeminiLiveVertexLLMService` now send it. Gemini treats a seed as best
  effort, so identical seeds usually but not always produce identical
  responses.
  (PR [#5366](https://github.com/pipecat-ai/pipecat/pull/5366))

- Fixed `GoogleLLMService` not applying its low-latency thinking defaults in
  `run_inference()`. Now both in-pipeline and `run_inference()` code paths
  build their request the same way. An explicit `thinking` setting still wins.
  (PR [#5368](https://github.com/pipecat-ai/pipecat/pull/5368))

- Fixed word-level TTS tracking recording the TTS-side text in the conversation
  context instead of the LLM's original text when a TTS service's
  word-timestamp events don't spell a word the way it was sent. This fixes
  cases where a provider strips diacritics (e.g. recording cafe instead of the
  LLM's café) and where text is closed out early while carrying synthesis tags,
  causing those tags (e.g. `</spell>`) to be recorded instead of the LLM's own
  pattern delimiters (e.g. `</card>`).
  (PR [#5370](https://github.com/pipecat-ai/pipecat/pull/5370))

- Pipecat now requires `pydantic>=2.13` on Python 3.14, where earlier pydantic
  releases ship no prebuilt wheels and must be compiled from source. Other
  Python versions are unaffected.
  (PR [#5375](https://github.com/pipecat-ai/pipecat/pull/5375))

- Fixed a negative TTFB being reported by an STT service that finalizes a
  segment on its own endpointing and then returns nothing for the final
  segment. The timeout path measured to that earlier transcript, which predates
  the speech it measured from, reporting the service as responding before it
  was asked. Such an utterance now reports no TTFB, and the service is named in
  a warning.
  (PR [#5384](https://github.com/pipecat-ai/pipecat/pull/5384))

- `FrameProcessorMetrics.stop_ttfb_metrics()` now refuses any measurement whose
  output predates its start, so a wall clock that steps backwards
  mid-measurement cannot put an impossible latency into the metrics stream.
  Processors can also call `cancel_ttfb_metrics()` to abandon a measurement
  whose response never arrived, rather than leaving it open for unrelated
  output to be measured against.
  (PR [#5384](https://github.com/pipecat-ai/pipecat/pull/5384))

- Fixed `MuteUntilFirstBotCompleteUserMuteStrategy` leaving the user muted for
  the rest of the call when the bot's first speaking turn failed. The strategy
  unmutes on `BotStoppedSpeakingFrame`, which a turn that produces no audio — a
  TTS failure, say — never emits, and a muted user can't prompt another turn to
  supply one. An `ErrorFrame` arriving before the bot starts speaking now
  releases the mute as well; errors after that point are ignored, since the
  output transport ends the turn on its own once the audio dries up.
  (PR [#5390](https://github.com/pipecat-ai/pipecat/pull/5390))

- Fixed `TavusTransport` ignoring its `bot_name` argument, so every Tavus bot
  joined the room named "Pipecat".
  (PR [#5392](https://github.com/pipecat-ai/pipecat/pull/5392))

- Fixed `language_code` being dropped on `eleven_v3` and
  `eleven_v3_conversational` in `ElevenLabsHttpTTSService`. The v3 models
  accept 74 languages — including Farsi, Pashto, and Sindhi, which no other
  ElevenLabs model covers — where `eleven_flash_v2_5` and `eleven_turbo_v2_5`
  accept 32. A language the selected model doesn't support is dropped with a
  warning rather than sent.
  (PR [#5398](https://github.com/pipecat-ai/pipecat/pull/5398))

- Fixed `LLMWorker` holding back frames that had nothing to do with a running
  tool. Frames queued while a `@tool` handler ran were deferred until it
  finished, which was meant for the handler's own output but caught everything:
  frames arriving over the bus and the worker's own lifecycle frames were held
  too. Deferral now applies only to frames queued from inside a handler, or
  from something it awaits. An application event handler that queues a frame
  while a tool happens to be running is no longer held behind it.
  (PR [#5399](https://github.com/pipecat-ai/pipecat/pull/5399))

- Fixed a worker acting on the same cancellation more than once. A worker
  receives more than one cancel on an ordinary shutdown, and only the pipeline
  frame was guarded, so each one was announced in the log again and propagated
  to every child, compounding down a tree of workers.
  (PR [#5399](https://github.com/pipecat-ai/pipecat/pull/5399))

- Fixed the deprecation registry scanner crashing on editor lock files. A lock
  symlink left beside a source file, such as Emacs's `.#module.py`, matched the
  scanner's glob but pointed at nothing readable, so pre-commit and
  `tests/test_deprecation_markers.py` failed for anyone with an unsaved buffer
  under `src/`.
  (PR [#5399](https://github.com/pipecat-ai/pipecat/pull/5399))

- Fixed `PipelineWorker.flush_pipeline()` reporting a drain that had not
  happened. The probe went straight to the pipeline, so it could overtake
  frames still waiting on the worker's push queue. It now queues behind them,
  while still bypassing `queue_frame` overrides such as the tool-call deferral.
  (PR [#5399](https://github.com/pipecat-ai/pipecat/pull/5399))

- Fixed `FilterIncompleteUserTurnStrategies` talking over the user when a `✓`
  (complete) verdict resolved after the user had already resumed speaking. Such
  a completion is stale: the user turn stays open, so no new turn start — and
  no interruption — could cut the bot off. `UserTurnCompletionLLMServiceMixin`
  now treats a `✓` that arrives while VAD hears the user as `○`, suppressing
  the response and re-arming the short re-prompt timeout.
  (PR [#5407](https://github.com/pipecat-ai/pipecat/pull/5407))

- Fixed a `SIGSEGV` in `libkrisp-audio-sdk` on pipeline teardown:
    - The Krisp VIVA SDK is now initialized once per process, instead of being
      destroyed whenever the last `KrispVivaFilter`, `KrispVivaVadAnalyzer`,
      `KrispVivaTurn`, or `KrispVivaIPUserTurnStartStrategy` reference was
      released. Its global state is shared by every Krisp session, so tearing it
      down while one of them was still processing audio left that session
      reading freed memory.
    - `KrispVivaVadAnalyzer.cleanup()` is now idempotent, so the two calls the
      pipeline makes per session no longer release one reference more than it
      acquired.
    - `KrispVivaFilter` and `KrispVivaVadAnalyzer` no longer release a reference
      their constructor never acquired.  (PR
      [#5411](https://github.com/pipecat-ai/pipecat/pull/5411))

- Fixed a flat ~0.5s delay on every user turn with STT services that do their
  own end-of-turn detection (`CartesiaTurnsSTTService`,
  `DeepgramFluxSTTService`, `DeepgramFluxSageMakerSTTService`, and
  `AssemblyAISTTService`/`SonioxSTTService` with
  `vad_force_turn_endpoint=False`). `ProposedUserStoppedSpeakingFrame` is now a
  `ControlFrame` rather than a `SystemFrame`, so it stays ordered behind the
  final `TranscriptionFrame` these services push ahead of it.
  `ExternalUserTurnStopStrategy` now has the text it needs to close the turn as
  soon as the proposal arrives, instead of waiting out its aggregation timer.
  (PR [#5423](https://github.com/pipecat-ai/pipecat/pull/5423))

- Fixed the WebSocket transports reporting a successful write for a frame that
  never went out, which pushed it downstream as though it had been delivered.
  (PR [#5424](https://github.com/pipecat-ai/pipecat/pull/5424))

- Fixed `BaseOutputTransport` hanging when a write to the transport never
  returns, for example when a client stops reading. The audio task stayed
  parked inside the write, so the bot went silent and the `EndFrame` never
  reached the end of the pipeline. Writes are now bounded by the new
  `TransportParams.audio_out_write_timeout_secs` (default 10s), and exceeding
  it leaves the transport unusable.
  (PR [#5424](https://github.com/pipecat-ai/pipecat/pull/5424))

- Fixed `CerebrasLLMService` dropping `max_tokens`, `frequency_penalty`,
  `presence_penalty` and `service_tier` from chat completion requests. All four
  are supported by the Cerebras API and are now sent.
  (PR [#5448](https://github.com/pipecat-ai/pipecat/pull/5448))

- Fixed the MoQ transport crashing when a fast peer's audio arrived before the
  input transport started: the session task died with `AttributeError:
  '_audio_in_queue'` and the bot stayed silent for the rest of the call. Audio
  received before `StartFrame` has created the audio queue is now dropped.
  (PR [#5451](https://github.com/pipecat-ai/pipecat/pull/5451))

### Performance

- Replaced the `pyloudnorm` dependency with `loudness`, which requires only
  `numpy`. This takes `scipy` out of the base install, where importing it
  accounted for roughly 690ms of cold start time. `import pipecat.audio.utils`
  now costs 0.25s instead of 0.95s.
  (PR [#5232](https://github.com/pipecat-ai/pipecat/pull/5232))

- ⚠️ `LLMContext` and `LLMService` no longer import the OpenAI SDK, so a
  pipeline that talks to another provider no longer loads it. `LLMContext`
  takes its "not provided" sentinel from Pipecat rather than the SDK, and
  `LLMService.adapter_class` defaults to `None`, resolving to
  `OpenAILLMAdapter` at construction. Bots on a non-OpenAI provider save about
  230ms of import.

  `LLMService.adapter_class` now reads as `None` rather than
  `OpenAILLMAdapter` when a subclass doesn't set it; use `get_llm_adapter()`
  for the resolved adapter instance. Code comparing `LLMContext`'s `NOT_GIVEN`
  against OpenAI's by identity should use `is_given()` instead.
  `LLMContext(tools=...)` and `set_tools()` likewise accept only Pipecat's
  `NOT_GIVEN` now, so callers passing OpenAI's should pass Pipecat's or omit
  the argument.
  (PR [#5253](https://github.com/pipecat-ai/pipecat/pull/5253))

- ⚠️ Added `pipecat.utils.types`, home of the `NOT_GIVEN` sentinel now shared
  by settings, `LLMContext` and anything else needing "this value was not
  provided", together with `is_given()` and `assert_given()`. Provider SDKs
  keep their own equivalents, translated at the adapter boundary.

  `NOT_GIVEN`, `NotGiven`, `is_given()` and `assert_given()` now come from
  `pipecat.utils.types` and are no longer importable from
  `pipecat.services.settings`. The private
  `pipecat.services.settings._NotGiven` is now the public
  `pipecat.utils.types.NotGiven`.

  The `is_given()` exported by the OpenAI and Anthropic adapters tests that
  SDK's sentinel rather than Pipecat's, and is now named `openai_is_given()`
  and `anthropic_is_given()` to keep the two apart.
  `pipecat.adapters.services.open_ai_adapter` also exports the translations
  that respell a context's values for the OpenAI SDK:
  `openai_from_llm_context_tools()`, `openai_from_llm_context_tool_choice()`
  and `openai_from_llm_standard_message()`. The tools one is shared by the Chat
  Completions and Responses adapters.
  (PR [#5253](https://github.com/pipecat-ai/pipecat/pull/5253))

- Cut Pipecat's import time roughly in half by loading heavy third-party
  dependencies on first use instead of at import. NLTK, which reaches
  `scikit-learn` and in turn `scipy` through its classifier backends, now loads
  inside `match_endofsentence()`, and `fastapi` is type-checking-only in
  `pipecat.runner.types` and `pipecat.runner.utils`. With `pyloudnorm` already
  replaced, `scipy` no longer loads at all for a typical bot. Importing the
  modules a voice bot uses drops from about 2.2s to about 1.0s.

  `PipelineWorker` warms NLTK on a background thread as the pipeline starts,
  so the opening bot turn doesn't pay the load either. The NLTK `punkt_tab`
  data check, which can hit the network, moves off module import to that
  warming. Images that bundle `punkt_tab` at build time, or set `NLTK_DATA` to
  a directory that has it, keep the warming off the network entirely.
  (PR [#5253](https://github.com/pipecat-ai/pipecat/pull/5253))

- Bots start faster. Services and transports connect while the pipeline is
  setting up rather than when the `StartFrame` arrives, and a pipeline sets up
  and cleans up its processors concurrently, so startup costs the slowest
  service rather than the sum of them all.
  (PR [#5316](https://github.com/pipecat-ai/pipecat/pull/5316))

## [1.7.0] - 2026-08-01

### Added

- Added token usage metrics to `AWSNovaSonicLLMService`, which now emits
  `LLMUsageMetricsData` from Nova Sonic's `usageEvent`. Each event reports its
  token delta, with speech and text tokens combined into `prompt_tokens` and
  `completion_tokens`, so metrics summed over a session match the session
  total.
  (PR [#4783](https://github.com/pipecat-ai/pipecat/pull/4783))

- Added an `http_client` parameter to `OpenAITTSService` and Whisper-based STT
  services (`BaseWhisperSTTService`, `OpenAISTTService`, `GroqSTTService`), so
  a custom `httpx.AsyncClient` — e.g. one with a raised request timeout for
  high-latency endpoints — can be used for API requests. Prefer
  `openai.DefaultAsyncHttpxClient`, which retains the OpenAI SDK's connection
  limits and redirect handling.
  (PR [#4941](https://github.com/pipecat-ai/pipecat/pull/4941))

- Added the ability for users to provide a local image to the LemonSlice
  transport to be used as the avatar image.
  (PR [#4977](https://github.com/pipecat-ai/pipecat/pull/4977))

- Added a `filter_background_audio` setting to
  `ElevenLabsRealtimeSTTService.Settings` so callers can have ElevenLabs
  suppress background and far-end audio before transcription, which reduces
  spurious partial transcripts and empty commits on noisy telephony audio. When
  set, it is forwarded as a connection query parameter regardless of commit
  strategy. Defaults to unset, which preserves ElevenLabs' default of no
  filtering.
  (PR [#5003](https://github.com/pipecat-ai/pipecat/pull/5003))

- Added STT usage metrics: every STT service reports usage as
  `STTUsageMetricsData`, carrying the client-measured seconds of audio
  submitted to the service (`audio_seconds`). Continuous services emit
  incrementally per final transcript with a flush on stop/cancel; segmented
  services emit per transcribed segment. Enabled with
  `enable_usage_metrics=True`; usage is forwarded to RTVI clients
  (`stt_usage`), logged by `MetricsLogObserver`, and attached to OpenTelemetry
  `stt` spans as `metrics.audio_seconds`.
  (PR [#5055](https://github.com/pipecat-ai/pipecat/pull/5055))

- Added inbound SIP DTMF support on the LiveKit transport via `on_dtmf_event`
  and `InputDTMFFrame`, so `DTMFAggregator` works with LiveKit SIP/PSTN calls.
  (PR [#5097](https://github.com/pipecat-ai/pipecat/pull/5097))

- Added `numerals` setting to `DeepgramFluxSTTSettings` to convert spoken
  numbers to numeral form in transcripts (e.g. "twenty three" → "23"). Enable
  with `numerals=True` in the settings; configured at connection time per
  Deepgram's Flux API.
  (PR [#5099](https://github.com/pipecat-ai/pipecat/pull/5099))

- Added `safety_settings` to `GoogleLLMService.Settings` (and
  `GoogleVertexLLMService.Settings`), exposing Gemini's content safety filters.
  Previously these could only be set by smuggling them through `extra`.

      ```python
      from google.genai.types import HarmBlockThreshold, HarmCategory,
  SafetySetting

      llm = GoogleLLMService(
          api_key=os.getenv("GOOGLE_API_KEY"),
          settings=GoogleLLMService.Settings(
              safety_settings=[
                  SafetySetting(
                      category=HarmCategory.HARM_CATEGORY_HATE_SPEECH,
                      threshold=HarmBlockThreshold.BLOCK_LOW_AND_ABOVE,
                  ),
              ],
          ),
      )
      ```

      Categories left unspecified keep the Gemini API defaults. The setting is
  runtime-updatable via `LLMUpdateSettingsFrame`, which also accepts plain
  dicts.
  (PR [#5109](https://github.com/pipecat-ai/pipecat/pull/5109))

- Added `language_codes` setting to `AssemblyAISTTSettings`, exposing the
  list-valued name of AssemblyAI's declared-language parameter and the one to
  prefer over the singular `language_code`, which stays supported and is
  ignored when both are set. It takes `Language` enums: a single language (e.g.
  `language_codes=[Language.ES]`) pins transcription to that language, while
  several (e.g. `language_codes=[Language.EN, Language.ES]`) steer toward that
  subset while keeping code-switching among them. Steering is prompt-based, so
  it applies to U3 Pro models only and isn't sent for other models; on U3 Pro a
  change applies to the live connection instead of reconnecting, and an empty
  list clears steering back to the model default. At most 10 distinct languages
  can be declared — regional variants resolve to their base code, so they
  collapse rather than counting twice. An over-long list raises at
  construction; an over-long runtime update is dropped with a warning, leaving
  the current steering in place.
  (PR [#5129](https://github.com/pipecat-ai/pipecat/pull/5129))

- Added `endpointing`, `keywords`, and `format` to
  `SmallestSTTService.Settings`. `endpointing` finalizes transcripts promptly
  on trailing silence, `keywords` boosts recognition of domain-specific
  words/phrases (e.g. `"Blackwell:2,NVIDIA:1"`), and `format` controls whether
  transcripts get punctuation and capitalization applied.
  (PR [#5137](https://github.com/pipecat-ai/pipecat/pull/5137))

- Added an opt-in `force_locale` setting to `AzureTTSService` and
  `AzureHttpTTSService` (`AzureTTSSettings`) that wraps synthesized text in
  SSML's `<lang xml:lang>` element, so multilingual voices (e.g.
  `en-US-EmmaMultilingualNeural`) speak in the configured locale/accent instead
  of auto-detecting one per segment. Defaults to `False`, leaving SSML output
  untouched unless enabled.
  (PR [#5154](https://github.com/pipecat-ai/pipecat/pull/5154))

- Added `reach_inactive_services` to `ServiceUpdateSettingsFrame`, and so to
  `LLMUpdateSettingsFrame`, `TTSUpdateSettingsFrame` and
  `STTUpdateSettingsFrame`. Set it when a settings update is provider-neutral
  and needs to survive a service switch, and every service a `ServiceSwitcher`
  manages applies it rather than the active one alone.

      ```python
      update = LLMUpdateSettingsFrame(
          delta=LLMSettings(temperature=0.2),
          reach_inactive_services=True,
      )
      await worker.queue_frames([update])
      ```

      It defaults to `False`, which suits values only one provider understands:
  a Cartesia voice id applied to a Deepgram TTS service would leave it unusable
  once it takes over. To configure one specific service, address the update to
  it with `service=` instead.
  (PR [#5155](https://github.com/pipecat-ai/pipecat/pull/5155))

- Added `keyterm` to `CartesiaSTTService.Settings` and
  `CartesiaTurnsSTTService.Settings`, biasing transcription toward
  domain-specific words and phrases such as product names and jargon.

      ```python
      stt = CartesiaTurnsSTTService(
          api_key=os.environ["CARTESIA_API_KEY"],
          settings=CartesiaTurnsSTTService.Settings(keyterm=["Pipecat", "Ink
  2"]),
      )
      ```

      Cartesia binds keyterms to a connection, so updating them with an
  `STTUpdateSettingsFrame` reconnects to apply them. Lists longer than
  Cartesia's limit of 100 keyterms or 1200 total characters are truncated with
  a warning. `CartesiaSTTService` sends keyterms only for ink-2 models, the
  only family Cartesia supports them on.
  (PR [#5168](https://github.com/pipecat-ai/pipecat/pull/5168))

- Added `PocketTTSService`, a local CPU-only text-to-speech service built on
  kyutai-labs' [pocket-tts](https://github.com/kyutai-labs/pocket-tts)
  streaming model. Supports English, French, German, Italian, Portuguese, and
  Spanish, predefined voices, and voice cloning from a wav file or `hf://`
  voice prompt. Install with `pip install "pipecat-ai[pocket-tts]"`.

      ```python
      from pipecat.services.pocket_tts.tts import PocketTTSService

      tts = PocketTTSService(settings=PocketTTSService.Settings(voice="alba"))
      ```
  (PR [#5170](https://github.com/pipecat-ai/pipecat/pull/5170))

### Changed

- `SmallestSTTService` now uses the Waves v4 STT endpoint
  (`/waves/v1/stt/live`) and sends `finalize` per utterance to keep the
  WebSocket session alive across turns.
  (PR [#4747](https://github.com/pipecat-ai/pipecat/pull/4747))

- The TeXML served by the development runner for Twilio and Telnyx no longer
  includes a trailing `<Pause length="40"/>`. `<Connect><Stream>` holds the
  call for the duration of the WebSocket session, so the call now hangs up as
  soon as the stream ends instead of lingering for another 40 seconds.
  (PR [#5135](https://github.com/pipecat-ai/pipecat/pull/5135))

- `SageMakerBidiClient` now relies on the SageMaker Runtime SDK's default auth
  configuration (SigV4 for the sagemaker service) instead of passing an
  equivalent explicit configuration, so future SDK auth changes are picked up
  automatically.
  (PR [#5144](https://github.com/pipecat-ai/pipecat/pull/5144))

- ⚠️ `TavusParams.audio_out_faster_than_realtime` now defaults to `True`. Bot
  audio is accumulated into 100ms chunks and sent to Tavus as fast as it is
  produced, giving the avatar a larger rendering buffer, instead of being paced
  to real playback time. Pipelines that need bot audio to arrive at roughly
  real time — for example when an `AudioBufferProcessor` downstream is
  recording the conversation — should now set
  `audio_out_faster_than_realtime=False` explicitly.
  (PR [#5162](https://github.com/pipecat-ai/pipecat/pull/5162))

- Websocket-based services now bound the websocket closing handshake at 2
  seconds instead of the `websockets` default of 10. A service disconnects
  while handling the `EndFrame`, before the frame continues downstream, so a
  peer that never acknowledges the close delayed pipeline shutdown by that much
  per service. The bound is configurable per service via the new
  `ws_close_timeout` argument on `WebsocketService`.
  (PR [#5164](https://github.com/pipecat-ai/pipecat/pull/5164))

- `CartesiaSTTService` now connects with `Cartesia-Version: 2026-03-01`, the
    version that supports `keyterm` and returns structured JSON errors. This
    brings it in line with `CartesiaTTSService` and `CartesiaTurnsSTTService`,
    which already use that version.
  (PR [#5168](https://github.com/pipecat-ai/pipecat/pull/5168))

- `SarvamLLMService` now defaults to `sarvam-105b`, the only chat model
  Sarvam's
    API still serves. `sarvam-30b`, `sarvam-30b-16k` and `sarvam-105b-32k` are
    rejected server-side, so they're no longer accepted as the `model` setting.
  (PR [#5177](https://github.com/pipecat-ai/pipecat/pull/5177))

- `BasetenLLMService` now defaults to `moonshotai/Kimi-K2.6`. Set
  `settings=BasetenLLMService.Settings(model=...)` to pin a different model.
  (PR [#5178](https://github.com/pipecat-ai/pipecat/pull/5178))

### Deprecated

- Deprecated `XTTSService`. Constructing one now emits a `DeprecationWarning`;
  it will be removed in 2.0.0. The [Coqui XTTS streaming
  server](https://github.com/coqui-ai/xtts-streaming-server) it connects to has
  been unmaintained since February 2024 and pins a commit of the discontinued
  `coqui-ai/TTS` library, and the XTTS-v2 model it serves is licensed for
  non-commercial use only.
      - `KokoroTTSService` and `PiperTTSService` are the maintained local TTS
  services. To stay on XTTS, the community-maintained
  [`pipecat-xtts-vllm`](https://docs.pipecat.ai/api-reference/server/services/tts/xtts-vllm)
  package targets a current XTTSv2 serving stack.
  (PR [#5131](https://github.com/pipecat-ai/pipecat/pull/5131))

- Deprecated `LLMSettings.filter_incomplete_user_turns` and
  `LLMSettings.user_turn_completion_config`, which will be removed in 2.0.0.
  Turn completion is configured through the user turn strategy: pass
  `user_turn_strategies=FilterIncompleteUserTurnStrategies()` to
  `LLMUserAggregatorParams`, with
  `FilterIncompleteUserTurnStrategies(config=...)` for a custom
  `UserTurnCompletionConfig`. Setting either field at construction now emits a
  `DeprecationWarning`; it never took effect there.
  (PR [#5157](https://github.com/pipecat-ai/pipecat/pull/5157))

### Fixed

- Fixed cached DTMF audio being returned at the wrong sample rate.
  `load_dtmf_audio()` keyed its in-memory cache by button only, so the first
  sample rate a button was loaded at was reused for every later request. The
  cache is now keyed by `(button, sample_rate)`.
  (PR [#4766](https://github.com/pipecat-ai/pipecat/pull/4766))

- Fixed an issue where sending several `STTUpdateSettingsFrame`s in quick
  succession to `DeepgramFluxSTTService` (or the SageMaker transport) could be
  rejected by Deepgram Flux with `Configure rejected: [unknown] Too many
  pending Configure messages.`. Configure messages are now sent one at a time,
  waiting for each to be acknowledged (bounded by a 5s timeout) before the next
  goes out, so bursts of mid-stream settings updates no longer trip the
  server-side pending-Configure cap.
  (PR [#4791](https://github.com/pipecat-ai/pipecat/pull/4791))

- Fixed `AWSBedrockLLMAdapter` raising `UnboundLocalError` when a user
  message's content list places the image before the text (e.g. a
  `UserImageRawFrame` followed by a prompt).
  (PR [#4796](https://github.com/pipecat-ai/pipecat/pull/4796))

- Fixed `FunctionCallUserMuteStrategy` raising `KeyError` on a
  `FunctionCallResultFrame` / `FunctionCallCancelFrame` whose tool call id it
  was not tracking, which aborted the user aggregator's handling of that frame
  and surfaced an `ErrorFrame` to the application. Such frames arrive when the
  LLM invokes the built-in `cancel_async_tool_call` (excluded from
  `FunctionCallsStartedFrame`, but it still emits a result), when an async tool
  reports intermediate updates (a result frame per update, plus the final one),
  and when a bus bridge re-delivers a result another worker already handled.
  Untracked tool call ids are now ignored.
  (PR [#4824](https://github.com/pipecat-ai/pipecat/pull/4824))

- Fixed `AWSBedrockLLMService` dropping all but the last tool call in streaming
  responses. Parallel tool calls arrive as separate Bedrock content blocks, but
  the accumulator overwrote the prior block on each `contentBlockStart`, so
  only the final call ran. Tool blocks are now keyed by `contentBlockIndex` and
  finalized on `contentBlockStop`, so every requested function call is
  dispatched.
  (PR [#4918](https://github.com/pipecat-ai/pipecat/pull/4918))

- Fixed `expand_currency` (used by `VoiceFormatter`) leaking the 3rd and later
  fractional digits of a currency amount onto the subunit word. `$5.500`
  expanded to `five dollars and fifty cents0` and `$3.567` to `three dollars
  and fifty-six cents7`, so TTS spoke the stray digit. The full fraction is now
  consumed and read to cent precision (e.g. `five dollars and fifty cents`),
  dropping sub-cent precision like a two-digit amount.
  (PR [#4969](https://github.com/pipecat-ai/pipecat/pull/4969))

- Fixed `TurnAnalyzerUserTurnStopStrategy` releasing user turns later than the
  configured STT `ttfs_p99_latency` budget. The safety-net timeout is now
  computed from an absolute deadline — the end of the user's speech plus
  `ttfs_p99_latency` — so time spent in end-of-turn analysis (e.g. ML
  turn-analyzer inference, typically 50–120ms) no longer extends the wait.
  (PR [#5007](https://github.com/pipecat-ai/pipecat/pull/5007))

- Fixed `TurnAnalyzerUserTurnStopStrategy` waiting out the STT safety-net timer
  when a transcript finalized while end-of-turn analysis was still running: the
  trigger conditions are re-checked as soon as the analyzer's verdict lands,
  releasing the turn immediately. With `wait_for_transcript=False`, the turn is
  likewise released directly on the analyzer's `COMPLETE` verdict.
  (PR [#5007](https://github.com/pipecat-ai/pipecat/pull/5007))

- Fixed `TaskManager.current_tasks()` undercounting live tasks, which left the
  dangling-task warnings logged at shutdown incomplete. The task registry was
  keyed by task name, and names are not unique — tasks started from the same
  method on the same object, such as the parallel function-call tasks, all
  share one name and evicted one another. The registry is now keyed by the task
  object.
  (PR [#5025](https://github.com/pipecat-ai/pipecat/pull/5025))

- Fixed services that send the system instruction as a separate parameter (e.g.
  `AnthropicLLMService`, `AWSBedrockLLMService`, `GoogleLLMService`)
  permanently rewriting a lone `system` message in the `LLMContext` to `user`.
  When a context's only message is a system message, it's converted to `user`
  so the provider isn't sent an empty conversation history; that conversion now
  happens on a copy, so the context keeps its `system` role and later turns —
  once more messages have accumulated — still send it as the system
  instruction.
  (PR [#5027](https://github.com/pipecat-ai/pipecat/pull/5027))

- Fixed `developer`-role messages being permanently rewritten to `user` in the
  `LLMContext` by OpenAI-compatible services that don't support the role (e.g.
  `QwenLLMService`, `OllamaLLMService`, `TogetherLLMService`). These services
  convert `developer` to `user` on the way out to the provider; that conversion
  now happens on a copy, so the stored conversation history keeps its original
  roles for logging, persistence, and any other service reading the same
  context.
  (PR [#5027](https://github.com/pipecat-ai/pipecat/pull/5027))

- Fixed `GeminiLiveLLMService` crashing with WebSocket error 1007 when seeding
  a session from a context containing tool calls (e.g. after a multi-agent
  handoff or a reconnect). Gemini Live only accepts text and media content, so
  the new `GeminiLiveLLMAdapter` summarizes function calls and their responses
  as text.

    ⚠️ `GeminiLiveLLMService` now uses `GeminiLiveLLMAdapter` rather than
  `GeminiLLMAdapter`, so any hand-crafted `LLMSpecificMessage`s aimed at it
  need `llm="gemini-live"` instead of `llm="google"`. We recommend you use
  `LLMService.create_llm_specific_message()`, which picks the right value for
  you.
  (PR [#5028](https://github.com/pipecat-ai/pipecat/pull/5028))

- Fixed `pipecat init` omitting the video avatar provider's extra from the
  generated project's `pipecat-ai` dependency. Choosing Tavus, HeyGen, or Simli
  produced a `bot.py` importing that service while `pyproject.toml` requested
  only the transport and cascade extras, so the scaffolded bot failed to import
  after a fresh `uv sync`.
  (PR [#5031](https://github.com/pipecat-ai/pipecat/pull/5031))

- Fixed `FilterIncompleteUserTurnStrategies` going silent after a tool call
  when the LLM spoke a `✓` acknowledgement (e.g. "Let me look that up, one
  moment.") in the same response as the call. The per-turn
  one-spoken-completion guard — which stops the acoustic detector's duplicate
  inferences from voicing the same turn twice — also swallowed the post-tool
  response. A function call now resets that guard, so the post-tool answer is
  spoken and the acknowledgement still plays in full.
  (PR [#5063](https://github.com/pipecat-ai/pipecat/pull/5063))

- Fixed `AssemblyAISTTService` raising a `ValueError` at construction when
  `prompt` and `keyterms_prompt` were set together on U3 Pro models
  (`u3-rt-pro`, `universal-3-5-pro`). AssemblyAI's server accepts the
  combination on these models, so both parameters are now sent on the streaming
  connect URL; the client-side mutual-exclusivity check still applies to older
  models.
  (PR [#5084](https://github.com/pipecat-ai/pipecat/pull/5084))

- Fixed word-timestamp and streamed-token TTS providers producing incorrect or
  duplicated text/transcript output when two audio contexts were in flight at
  the same time (e.g. back-to-back `TTSSpeakFrame` utterances on a websocket
  TTS service). Previously the `AggregatedFrameSequencer` tracked
  pending-sentence and force-complete state globally, so a word event or
  force-complete for one context could bleed into another; state is now tracked
  per context. In streaming (token) mode, sentence-boundary frames were also
  being emitted twice per sentence; only one is now emitted.
  (PR [#5098](https://github.com/pipecat-ai/pipecat/pull/5098))

- Fixed streamed-token TTS providers with no word-timestamp support (e.g.
  `DeepgramFluxTTSService` in its default `TextAggregationMode.TOKEN`
  configuration) tracking each LLM token as its own spoken unit instead of
  grouping tokens into sentences, and never sending the RTVI "new segment"
  bot-output event. Previously the `AggregatedFrameSequencer` only grouped
  streamed tokens into sentences when the TTS service also supported word
  timestamps; providers without word timestamps now get the same per-sentence
  grouping, completing each sentence's slot as soon as it is confirmed (there
  is no word-timestamp signal to complete it progressively).
  (PR [#5102](https://github.com/pipecat-ai/pipecat/pull/5102))

- Fixed `ElevenLabsTTSService` pushing a spurious `ErrorFrame` (which could
  e.g. trigger an unwanted `ServiceSwitcherStrategyFailover` switch) when the
  server closed the websocket first during an intentional disconnect.
  (PR [#5103](https://github.com/pipecat-ai/pipecat/pull/5103))

- Fixed TTS services with `pause_frame_processing=True` (e.g. ElevenLabs,
  Deepgram, Azure, Rime, Inworld) permanently hanging the pipeline when the
  expected resume signal never arrived — for example, a provider reporting a
  successful completion with zero audio (a quota-exhausted key), or a resume
  racing ahead of the pause taking effect. The stuck pause blocked the terminal
  `EndFrame` from draining, which in turn blocked pipeline teardown,
  `PipelineRunner.run()`/`WorkerRunner` returning, and any cleanup that runs
  afterward. `TTSService` now starts a watchdog whenever it pauses frame
  processing and force-resumes (reporting a non-fatal `ErrorFrame`) if no
  `BotStartedSpeakingFrame` confirms audio within the new
  `pause_watchdog_timeout_s` parameter (default 3.0 seconds).
  (PR [#5106](https://github.com/pipecat-ai/pipecat/pull/5106))

- Fixed `LLMFullResponseEndFrame` being silently dropped for every turn after
  the first when two turns' audio contexts were in flight concurrently. The end
  frame is now looked up per audio context instead of being gated by a single
  shared "response started" flag, which only ever fired for whichever turn's
  audio happened to finish first.
  (PR [#5108](https://github.com/pipecat-ai/pipecat/pull/5108))

- Fixed `LLMFullResponseStartFrame`/`LLMFullResponseEndFrame` arriving out of
  order at downstream processors (e.g. `LLMAssistantAggregator`) when a new LLM
  turn started while the previous turn's audio was still playing — reachable on
  TTS services that don't pause frame processing during synthesis (e.g.
  `CartesiaTTSService`), where a second, overlapping turn is possible.
  `LLMFullResponseStartFrame` is now routed through the same serialization
  queue as the rest of a turn's frames, so it can no longer be pushed ahead of
  an earlier turn's still-draining audio context.
  (PR [#5108](https://github.com/pipecat-ai/pipecat/pull/5108))

- Fixed `SmallWebRTCConnection`'s outgoing app-message buffer never being
  flushed for data channels created by the remote peer. aiortc surfaces such
  channels already open, so the "open" event the flush was wired to never fired
  and buffered messages were stuck until disconnect; the buffer now also
  flushes when a channel arrives already open.
  (PR [#5112](https://github.com/pipecat-ai/pipecat/pull/5112))

- Fixed `SmallWebRTCTransport` silently dropping app messages sent before the
  peer connection was established — e.g. the RTVI `user-mute-started` emitted
  by `MuteUntilFirstBotCompleteUserMuteStrategy` at pipeline start, which left
  clients that connect slower than the pipeline starts (real networks; loopback
  usually wins the race) never showing the greeting-hold mute. Messages sent
  before the data channel is open are now buffered and delivered, in order,
  once it opens; messages sent while the connection is closing are discarded
  with a debug log.
  (PR [#5112](https://github.com/pipecat-ai/pipecat/pull/5112))

- Fixed `DeepgramSTTService` retrying forever with no error notification when
  the WebSocket connection kept failing (e.g. an invalid API key). A 4xx error
  from Deepgram now stops the retry loop immediately, and any other connection
  failure now reports a `push_error` on every attempt and gives up after 3
  consecutive failures within 5 seconds of connecting, backing off
  exponentially between attempts.
  (PR [#5113](https://github.com/pipecat-ai/pipecat/pull/5113))

- Fixed `SmallWebRTCTransport` input audio ignoring a configured
  `audio_in_channels=2`: the resampler was hardcoded to downmix to mono
  regardless of the setting, so frames were labeled as stereo while containing
  mono data. The resampler now targets the configured channel layout.
  (PR [#5115](https://github.com/pipecat-ai/pipecat/pull/5115))

- Fixed `SmallWebRTCTransport` input audio being read as half-speed garbage
  when `audio_in_sample_rate` matches the wire rate (48000): stereo frames
  bypassed the mono-downmixing resampler and their interleaved bytes were
  labeled mono.
  (PR [#5115](https://github.com/pipecat-ai/pipecat/pull/5115))

- Fixed a Pipecat CLI issue where one broken plugin took down the whole CLI.
  Plugins are imported on every invocation, so a bad install (missing
  transitive dependency, version conflict, stale wheel) made every command —
  `pipecat init` included — exit with a message saying the `cli` extra wasn't
  installed, which was not the problem. Plugins now load in isolation: a
  failure is reported on stderr and skipped, and the rest of the CLI carries
  on. A plugin that isn't a `typer.Typer` is rejected the same way instead of
  failing later with an opaque `AttributeError`.
  (PR [#5121](https://github.com/pipecat-ai/pipecat/pull/5121))

- Fixed a Pipecat CLI issue where the hint for enabling an optional sub-CLI
  silently uninstalled other plugins. `uv tool install --with` replaces the
  tool environment rather than adding to it, so a hint naming only the missing
  plugin dropped every other one. The hint now repeats every installed plugin
  package.
  (PR [#5121](https://github.com/pipecat-ai/pipecat/pull/5121))

- Fixed `SmallWebRTCTransport` PATCH `/api/offer` requests crashing with an
  `AssertionError` when the browser sends an RFC 8840 end-of-candidates marker
  (an empty candidate string) for a media line, as Firefox does on every
  connection. The marker is now forwarded to aiortc as `None` instead of being
  parsed as a regular candidate.
  (PR [#5126](https://github.com/pipecat-ai/pipecat/pull/5126))

- Fixed word-by-word context attribution breaking after punctuation for TTS
  services that report word timestamps (e.g. `InworldTTSService`), which
  duplicated the affected sentence in the assistant context. When a provider's
  word-timestamp events don't repeat punctuation attached to the previous word
  — reporting `"Yeah"` then `"I"` for `"Yeah, I can do that."` — every word
  after the punctuation failed to match its slot, was emitted as an untracked
  passthrough frame without its original-text attribution, and the unmatched
  remainder was re-emitted when the audio context ended.
  (PR [#5128](https://github.com/pipecat-ai/pipecat/pull/5128))

- Fixed punctuation being duplicated in the assistant context when a TTS
  service reports it with the following word (`", I"`) rather than the
  preceding one (`"Yeah,"`), which produced `"Yeah, , I can do that."`.
  Punctuation trailing a word is attributed to that word, so the repeat is now
  dropped from the following word instead of discarding that word's
  original-text attribution.
  (PR [#5128](https://github.com/pipecat-ai/pipecat/pull/5128))

- Fixed `AzureTTSService` and `AzureHttpTTSService` crashing at startup when
  `Settings(language=None)` was used: the synthesis language is no longer
  assigned to the Azure SDK (which rejects `None`), letting the service default
  apply.
  (PR [#5144](https://github.com/pipecat-ai/pipecat/pull/5144))

- Fixed `DailyTransportClient.add_custom_video_track()` crashing with an
  `AttributeError`: the video track factory was invoked without `await`, so the
  SDK received a coroutine instead of a track. Joining with an already-released
  client now reports a join error instead of raising a `TypeError`.
  (PR [#5144](https://github.com/pipecat-ai/pipecat/pull/5144))

- Fixed an `InvalidSsmlException` error in `AWSPollyTTSService` when the text
  to synthesize contained XML-reserved characters such as `&` or `<`. The text
  is now escaped before being embedded in the SSML request.
  (PR [#5144](https://github.com/pipecat-ai/pipecat/pull/5144))

- Fixed `NvidiaSegmentedSTTService` failing with gRPC `NOT_FOUND` against
  NVIDIA's hosted endpoint: the default NVCF `function_id` for `canary-1b-asr`
  now points at NVIDIA's current deployment.
  (PR [#5144](https://github.com/pipecat-ai/pipecat/pull/5144))

- Fixed responses to `LLMMessagesAppendFrame(..., run_llm=True)` being silently
  dropped when `filter_incomplete_user_turns` is enabled and the bot had
  already responded in the current user turn. The response was generated but
  never spoken, with nothing logged — most visibly breaking user-idle
  re-prompts, where an escalation ladder ended the session with an unspoken
  goodbye.
  (PR [#5146](https://github.com/pipecat-ai/pipecat/pull/5146))

- Fixed `ServiceSwitcher` (and therefore `LLMSwitcher`) delivering
  `ServiceUpdateSettingsFrame` only to the active service, so an update
  addressed to a specific service — `LLMUpdateSettingsFrame(service=...)`,
  `TTSUpdateSettingsFrame(service=...)` — never reached its target unless that
  target happened to be active. An addressed update now reaches its service
  whether or not it is the active one. Updates with no target still apply to
  the active service alone, since settings values are often specific to one
  provider; mark one `reach_inactive_services` to send it to every service the
  switcher manages.
  (PR [#5155](https://github.com/pipecat-ai/pipecat/pull/5155))

- Fixed `LLMTurnCompletionUserTurnStopStrategy` configuring only the LLM active
  at pipeline start when its LLMs sit behind an `LLMSwitcher`. Once a switch
  made another LLM active, that LLM emitted none of the turn-completion markers
  the strategy waits on, and every user turn was closed by the
  `user_turn_stop_timeout` watchdog instead. The startup update now reaches
  every LLM the switcher manages.
  (PR [#5155](https://github.com/pipecat-ai/pipecat/pull/5155))

- Fixed `TurnAnalyzerUserTurnStopStrategy` ending the user turn on every
  finalized transcript when the turn was started by a transcript rather than by
  a VAD frame — the case for `MinWordsUserTurnStartStrategy`, and for
  `TranscriptionUserTurnStartStrategy` in the default strategy set. One
  utterance split by the STT endpointer into several finalized transcripts
  produced an LLM inference per fragment, each landing as its own `role:
  "user"` message, so the bot replied to partial phrases.
  (PR [#5159](https://github.com/pipecat-ai/pipecat/pull/5159))

- Fixed `total_tokens` omitting cached input tokens on the Anthropic and AWS
  Bedrock LLM services. Both report their input count net of the prompt cache,
  so cache reads and cache writes were left out of the total — a large
  undercount for a voice agent, where the cached prefix comes to dominate the
  input within a few turns. `total_tokens` is now the gross figure on both,
  matching what the OpenAI and Google services already reported, and on Bedrock
  a usage report carrying only cache activity is no longer skipped. The
  per-field breakdown is unchanged; read `prompt_tokens` for the net input
  figure.
  (PR [#5163](https://github.com/pipecat-ai/pipecat/pull/5163))

- Fixed `DTMFAggregator` emitting flushed `TranscriptionFrame`s with
  `finalized=False`, which stalled user-turn stop strategies waiting for a
  final transcription.
  (PR [#5172](https://github.com/pipecat-ai/pipecat/pull/5172))

- Fixed `UserIdleController` ignoring `UserIdleTimeoutUpdateFrame` until the
  next re-arm: a running idle timer now restarts with the new duration, and
  enabling a positive timeout while waiting for the user to speak arms the
  timer immediately.
  (PR [#5173](https://github.com/pipecat-ai/pipecat/pull/5173))

- Fixed `TTSService` leaving frame processing permanently paused when a text
  filter strips the text to empty and `pause_frame_processing` is enabled.
  (PR [#5174](https://github.com/pipecat-ai/pipecat/pull/5174))

- Fixed `NovitaLLMService` emitting a token-usage `MetricsFrame` (and a debug
  log line) for every streamed chunk instead of one per completion, which
  over-counted a single turn dozens of times over for anything aggregating
  those metrics. Novita reports a cumulative usage snapshot on each chunk; the
  service now collapses them into one total reported when the completion
  finishes, including when it is interrupted mid-stream.
  (PR [#5180](https://github.com/pipecat-ai/pipecat/pull/5180))

- Fixed `PerplexityLLMService` and `NvidiaLLMService` discarding the cache-read
  and reasoning token counts their providers report. Both now pass the whole
  usage snapshot through, so `cache_read_input_tokens` and `reasoning_tokens`
  reach the token-usage `MetricsFrame` alongside the prompt and completion
  counts.
  (PR [#5180](https://github.com/pipecat-ai/pipecat/pull/5180))

- Fixed a `KeyError: 'role'` crash in `AnthropicLLMService` when using a model
  that thinks by default, such as Claude Sonnet 5 or Claude Opus 5. The first
  tool call failed, and so did every later inference on that context. These
  models return thinking blocks whose text is empty and whose signature carries
  the reasoning; such blocks are now preserved and passed back to Anthropic, as
  required within a tool-use turn. A thought carrying no signature has no valid
  thinking-block form and is skipped.
  (PR [#5183](https://github.com/pipecat-ai/pipecat/pull/5183))

## [1.6.0] - 2026-07-21

### Added

- Added `MOQTransport`, a Media over QUIC (MoQ) transport that gives bots a
  bidirectional, low-latency audio + RTVI channel over QUIC instead of WebRTC
  or WebSockets. Install with `pip install pipecat-ai[moq]` and see
  `examples/transports/transports-moq.py`.
    - The bot runs as its own MoQ server (`serve=True`) and accepts the
  browser's direct connection, removing the need for a separate `moq-relay`
  process in local dev; client mode (dialing an external relay) is wired up but
  not yet enabled.
    - Audio rides a single Opus track; RTVI messages (including the transcript)
  ride a compressed, ordered JSON stream track, so MoQ is on par with the Daily
  and WebSocket transports for RTVI support.
    - The development runner (`pipecat.runner.run`) gained `--moq-serve`,
  `--moq-bind`, `--moq-tls-generate`/`--moq-tls-cert`/`--moq-tls-key` and
  related flags to configure the MoQ server and TLS for local dev.
  (PR [#4629](https://github.com/pipecat-ai/pipecat/pull/4629))

- Added `reasoning` support to `OpenAIResponsesLLMService` and
  `OpenAIResponsesHttpLLMService`. Set `settings.reasoning` to an
  `OpenAIResponsesLLMService.ReasoningConfig(effort=..., summary=...)` to
  control reasoning depth and, optionally, request a summary of the model's
  thinking. Summaries are surfaced the same way as Anthropic/Gemini thinking —
  as thought frames and the `on_assistant_thought` event. Reasoning is only
  supported by reasoning-capable models (the gpt-5.x series and the o-series);
  the default model, `gpt-4.1`, does not reason — see OpenAI's [reasoning
  guide](https://platform.openai.com/docs/guides/reasoning) to pick a model.

    The model's encrypted reasoning is captured and sent back on subsequent
  turns automatically, preserving reasoning context across the conversation
  (and, with function calling, across tool-call turns). See
  `examples/thinking/thinking-openai-responses.py` (plus the `-http` and
  `-functions-` variants).

    When `reasoning` is not configured, the mainline gpt series from gpt-5
  onward defaults to `effort="none"` (reasoning disabled) to keep latency low
  for real-time voice — mirroring how the Gemini service disables thinking by
  default — while every other model is left at its provider default.
  Conversely, if you configure `reasoning` on a model known not to support it
  (e.g. `gpt-4.1`), the service logs a clear error up front instead of leaving
  you to decipher the raw API failure.
  (PR [#4933](https://github.com/pipecat-ai/pipecat/pull/4933))

- Added `NO_RESPONSE` to Pipecat Flows: a consolidated function can return
  `(result, NO_RESPONSE)` to finish the function call without transitioning to
  a new node or running the LLM. The next response can then be triggered by the
  next user utterance, or programmatically another way.
  (PR [#4995](https://github.com/pipecat-ai/pipecat/pull/4995))

- Added `absent: true` to eval scenario expectations: the expectation passes
  only when no event of the given type arrives within the `within_ms` budget,
  and fails as soon as one does. Useful for duplicate-output regressions, e.g.
  asserting a bot responds exactly once after a multi-worker handoff.
  (PR [#4995](https://github.com/pipecat-ai/pipecat/pull/4995))

- Added `CrusoeLLMService`, an OpenAI-compatible LLM service for Crusoe Cloud's
  Managed Inference API.
  (PR [#5024](https://github.com/pipecat-ai/pipecat/pull/5024))

- Added audio token usage to `LLMTokenUsage` for cost attribution with realtime
  models: optional `input_audio_tokens`, `output_audio_tokens`, and
  `cache_read_input_audio_tokens` fields. `OpenAIRealtimeLLMService` (and Azure
  realtime) now populates them from the Realtime API's `response.done` usage
  details, and they flow through the usage debug logs, RTVI client metrics
  (only present when populated), and OTel span attributes
  (`gen_ai.usage.audio.input_tokens`, `gen_ai.usage.audio.output_tokens`,
  `gen_ai.usage.audio.cache_read.input_tokens`).
  (PR [#5050](https://github.com/pipecat-ai/pipecat/pull/5050))

- Added audio token usage capture to `GeminiLiveLLMService`: the AUDIO entries
  from `usage_metadata`'s per-modality breakdowns now populate
  `LLMTokenUsage`'s `input_audio_tokens`, `output_audio_tokens`, and
  `cache_read_input_audio_tokens`, flowing through usage logs, RTVI client
  metrics, and the `gen_ai.usage.audio.*` span attributes. Absent modalities
  are reported as unset rather than zero, and text tokens are never derived
  from totals (Gemini's modality details don't always sum to
  `prompt_token_count`). Gemini Live spans also now include cached and
  reasoning token counts, which the metrics path reported but spans were
  missing.
  (PR [#5052](https://github.com/pipecat-ai/pipecat/pull/5052))

- The development runner now prints a bordered startup banner flagging it as
  development-only, with a link to the deployment docs for running bots locally
  and in production.
  (PR [#5060](https://github.com/pipecat-ai/pipecat/pull/5060))

- Added `DeepgramFluxTTSService`, a websocket TTS service for Deepgram's Flux
  TTS (early access) at `wss://api.deepgram.com/v2/speak`. LLM tokens are
  streamed straight to the server as they arrive (`TextAggregationMode.TOKEN`,
  the default for this service; pass
  `text_aggregation_mode=TextAggregationMode.SENTENCE` to aggregate sentences
  instead) and each bot response is synthesized as a discrete turn, with
  prosody carried across turns on a single connection. Flux does not yet
  provide a way to cancel the active turn, so interruptions reconnect the
  websocket; `examples/voice/voice-deepgram-flux.py` is now an all-Flux bot
  (Flux STT + Flux TTS).
  (PR [#5067](https://github.com/pipecat-ai/pipecat/pull/5067))

- Added `BasetenLLMService`, an OpenAI-compatible LLM service for Baseten's
  Model APIs and dedicated deployments.

      Defaults to Baseten's serverless Model APIs endpoint, which serves
  open-weights models including GLM, Kimi, DeepSeek, Nemotron, and gpt-oss. To
  use a model running on your own dedicated GPUs, pass that deployment's
  `/sync/v1` URL as `base_url` and set `model` to its served model name:

      ```python
      llm = BasetenLLMService(
          api_key=os.getenv("BASETEN_API_KEY"),
          base_url=deployment_url,
          settings=BasetenLLMService.Settings(
              model="Qwen/Qwen2.5-3B-Instruct",
          ),
      )
      ```
  (PR [#5077](https://github.com/pipecat-ai/pipecat/pull/5077))

- `DailyTransport` now broadcasts an `STTMetadataFrame` with Deepgram's TTFS
  P99 latency when `transcription_enabled=True` and transcription starts
  successfully, matching standalone STT services. Downstream consumers like
  `LLMUserAggregator` and the user-turn-stop strategies now use the correct STT
  latency instead of falling back to defaults.
  (PR [#5088](https://github.com/pipecat-ai/pipecat/pull/5088))

### Changed

- Changed the default ElevenLabs TTS model from `eleven_turbo_v2_5` to
  `eleven_flash_v2_5` in `ElevenLabsTTSService` and `ElevenLabsHttpTTSService`,
  since `eleven_turbo_v2_5` is now deprecated by ElevenLabs. This only affects
  users who don't explicitly set a `model`.
  (PR [#4999](https://github.com/pipecat-ai/pipecat/pull/4999))

- Bumped the minimum `nltk` version to 3.10.0.
  (PR [#5019](https://github.com/pipecat-ai/pipecat/pull/5019))

- ⚠️ The RTVI `dtmf` client message now carries `buttons` — a list of keypad
  entries, e.g. `{"type": "dtmf", "data": {"buttons": ["1", "2", "#"]}}` — so a
  single message can press a whole key sequence. The previous single-key
  `button` field is no longer accepted, and `RTVI.PROTOCOL_VERSION` is now
  `2.1.0`. The `RTVIProcessor` pushes one `InputDTMFFrame` per key, in order,
  so downstream DTMF handling (e.g. a `DTMFAggregator`) behaves exactly as
  before.
  (PR [#5030](https://github.com/pipecat-ai/pipecat/pull/5030))

- Removed the `pyyaml-include` dependency (GPL-3.0), replacing it with a small
  built-in `!include` constructor for eval scenarios.
  (PR [#5037](https://github.com/pipecat-ai/pipecat/pull/5037))

- Updated tracing span attributes to the current OpenTelemetry GenAI semantic
  conventions: `gen_ai.provider.name` is now `azure.ai.openai` (was
  `az.ai.openai`) for `AzureLLMService`, `x_ai` (was `xai`) for
  `GrokLLMService`, and `mistral_ai` (was `mistral`) for `MistralLLMService`;
  reasoning token usage is now reported as
  `gen_ai.usage.reasoning.output_tokens` (was `gen_ai.usage.reasoning_tokens`).
  Update any dashboards or queries filtering on the old values.
  (PR [#5047](https://github.com/pipecat-ai/pipecat/pull/5047))

- Changed OpenAI Realtime `llm_response` span attributes to standard OTel GenAI
  names: `tokens.prompt`/`tokens.completion`/`tokens.total` are now
  `gen_ai.usage.input_tokens`/`gen_ai.usage.output_tokens`, plus the new
  cached/audio breakdown attributes. Update any dashboards or queries filtering
  on the old `tokens.*` names.
  (PR [#5050](https://github.com/pipecat-ai/pipecat/pull/5050))

- Changed Gemini Live `llm_response` span attributes: the non-standard
  `tokens.prompt`/`tokens.completion`/`tokens.total` were removed in favor of
  the standard `gen_ai.usage.input_tokens`/`gen_ai.usage.output_tokens`
  attributes already present on the same spans. Update any dashboards or
  queries filtering on the old `tokens.*` names.
  (PR [#5052](https://github.com/pipecat-ai/pipecat/pull/5052))

- Updated the `runner` extra to require `pipecat-ai-prebuilt>=1.0.4`.
  (PR [#5061](https://github.com/pipecat-ai/pipecat/pull/5061))

- Updated the runner extra to require `pipecat-ai-prebuilt>=1.0.5` to add
  support for the MoQ transport.
  (PR [#5073](https://github.com/pipecat-ai/pipecat/pull/5073))

- `TTSService` now logs `Generating TTS [text]` itself, just before invoking
  `run_tts`: at debug level in sentence aggregation mode and at trace level
  when streaming tokens (`TextAggregationMode.TOKEN`), where the accumulated
  turn text is already logged at debug level at flush time. The duplicate
  per-service logs — which logged every token at debug level in token mode —
  were removed from all TTS services.
  (PR [#5079](https://github.com/pipecat-ai/pipecat/pull/5079))

### Deprecated

- Deprecated `reset()` on `BaseUserTurnStartStrategy` and
  `BaseUserTurnStopStrategy`. Strategy "reset" logic should now happen through
  the new `handle_user_turn_started()` / `handle_user_turn_stopped()` lifecycle
  callbacks. For backward compatibility for strategy implementers who haven't
  implemented the new methods yet: by default, both start and stop strategies'
  `handle_user_turn_started()` invoke `reset()`, and stop strategies'
  `handle_user_turn_stopped()` invoke `reset()`. Because of this, a custom
  strategy that extends the base class and still overrides `reset()` keeps
  working. Overriding `reset()` emits a `DeprecationWarning` when the class is
  defined. To be removed in 2.0.0.
  (PR [#4967](https://github.com/pipecat-ai/pipecat/pull/4967))

- Deprecated the `webrtc` extra's dependency on OpenCV.
  `opencv-python-headless` will be removed from the `webrtc` extra in 2.0.0.
  Video pipelines using `SmallWebRTCTransport` should start installing the new
  `webrtc-video` extra (`pipecat-ai[webrtc-video]`) instead.
  (PR [#4978](https://github.com/pipecat-ai/pipecat/pull/4978))

- Deprecated `PronunciationDictionaryLocator` and the
  `pronunciation_dictionary_locators` parameter on `ElevenLabsTTSService` and
  `ElevenLabsHttpTTSService`. Pronunciation dictionary substitutions can
  rewrite the spoken words in ways that no longer match the text sent to
  synthesis, which breaks the alignment-based word-completion tracking used to
  attribute spoken text back to the conversation context. Use the
  `text_transforms` parameter with `replace_text` (added in 1.5.0) instead —
  those transforms run client-side and are tracked correctly. Will be removed
  in 2.0.0.
  (PR [#4991](https://github.com/pipecat-ai/pipecat/pull/4991))

### Fixed

- Fixed Gemini thinking-mode parallel tool calls: the adapter now groups the
  matching `function_response` messages into a single turn alongside the merged
  `function_call` messages, so the response count matches the call count. The
  previous mismatch was rejected with a `400` by the Vertex AI endpoint (the
  Gemini Developer API currently tolerates it).
  (PR [#4103](https://github.com/pipecat-ai/pipecat/pull/4103))

- Fixed `RTVIProcessor` echoing the client's declared major version back in
  `bot-ready` when connected to a legacy 1.x RTVI client, so the client's
  aggregator picks the code path that matches the wire format the server
  observer is actually sending.
  (PR [#4629](https://github.com/pipecat-ai/pipecat/pull/4629))

- Fixed `BaseWorker.send_job_stream_end` not removing the finished job from
  `_active_jobs`. Streaming jobs stayed "active" after the stream ended, so a
  cancel or update that raced in afterwards was still processed (firing
  `on_job_cancelled` and sending a `CANCELLED` response for an
  already-completed job) and the job leaked until worker shutdown. It now
  clears the job like `send_job_response` does.
  (PR [#4955](https://github.com/pipecat-ai/pipecat/pull/4955))

- Fixed `normalize_dates` rendering the month name via the locale-sensitive
  `strftime("%B")`, which produced a mixed-language spoken date (e.g. "Mai
  10th, two thousand and twenty-three") when the process locale was
  non-English. Month names are now always English, matching the rest of the
  spoken date.
  (PR [#4956](https://github.com/pipecat-ai/pipecat/pull/4956))

- Fixed `SingleClientWebsocketServerTransport` closing the client connection as
  soon as an `EndFrame` passed the input transport, cutting off output still
  being flushed downstream, such as a farewell spoken via `TTSSpeakFrame` or
  Pipecat Flows' `end_conversation` action `text`. The shared server is now
  reference-counted by the input and output transports and drained only once
  the last side has stopped, so the output finishes writing before the
  connection closes.
  (PR [#4964](https://github.com/pipecat-ai/pipecat/pull/4964))

- Fixed `TurnAnalyzerUserTurnStopStrategy` keeping stale speech state across
  externally-ended user turns. When a turn ended by any path other than the
  analyzer's own COMPLETE (a forced or external stop, or the stop watchdog
  timeout) while a mute strategy held audio back, the smart turn analyzer
  stayed frozen mid-speech and its `stop_secs` silence timer later fired a
  phantom end-of-turn during the bot's response (spurious
  `on_user_turn_inference_triggered` / `on_user_turn_stopped`, and with
  realtime services like Gemini Live a stray `activity_end` that could drop the
  user's next utterance). `UserTurnController` now notifies every stop strategy
  that the turn ended via a `handle_user_turn_stopped()` callback — on every
  stop path, never at turn start — and `TurnAnalyzerUserTurnStopStrategy`
  clears its analyzer there. The clear can't happen in the strategy's shared
  per-turn reset, which also runs at turn start, because that would drop the
  continuously-fed `pre_speech_ms` buffer; the buffer refills from incoming
  audio, so the next turn regains its full pre-speech context within
  `pre_speech_ms` of the clear, and only a user restarting within that window
  (about half a second) sees one prediction with a shortened lead-in.
  (PR [#4967](https://github.com/pipecat-ai/pipecat/pull/4967))

- Fixed `DeepgramSTTService` not reporting STT TTFB metrics when the finalize
  response has an empty transcript. When Deepgram's own endpointing fires
  before the `Finalize` command, the transcript is sent in a regular `is_final`
  response and the subsequent `from_finalize=True` response arrives with empty
  text. `confirm_finalize()` was inside the `len(transcript) > 0` guard and was
  never called, so `finalized` was never set and `stop_ttfb_metrics()` fell
  through to the 2-second timeout — arriving after `BotStartedSpeakingFrame`
  and missing the `UserBotLatencyObserver` window.
  (PR [#4973](https://github.com/pipecat-ai/pipecat/pull/4973))

- Fixed TTS text replacements (`text_transforms`/`text_filters`, e.g.
  `replace_text()`) corrupting the conversation context in word-timestamp mode
  (e.g. `CartesiaTTSService`) when a replacement split one word into several
  (`"BODYPUMP"` → `"body pump"`), changed only case (`"SQL"` → `"sql"`), or
  changed only the connector between words (`"BODYPUMP"` → `"body-pump"`).
  Previously the spoken (replaced) text could leak into the LLM context instead
  of the original text, or be silently dropped. Acronym letter-spacing (e.g.
  `"API"` → `"A P I"`) and inline IPA pronunciation tags had the same
  underlying issue and are fixed as well.
  (PR [#4976](https://github.com/pipecat-ai/pipecat/pull/4976))

- Fixed `SmallWebRTCTransport` importing OpenCV even for audio-only pipelines.
  `cv2` is now imported lazily, only when a non-RGB video frame actually needs
  converting. The `webrtc` extra's OpenCV dependency also switched from
  `opencv-python` to `opencv-python-headless`, avoiding GUI system library
  requirements.
  (PR [#4978](https://github.com/pipecat-ai/pipecat/pull/4978))

- Fixed `SpeechTimeoutUserTurnStopStrategy` ending the user turn mid-utterance
  when a new turn's `reset()` cleared VAD speaking state while the user was
  still talking without an intervening `VADUserStoppedSpeakingFrame`. A
  finalized transcript for a mid-utterance segment (as streaming STT services
  emit) was then treated as a standalone utterance with no active VAD
  reference, and `user_speech_timeout` falsely ended the turn. `reset()` now
  leaves VAD speaking state untouched, since it reflects live physical VAD
  state rather than turn-scoped bookkeeping.
  (PR [#4983](https://github.com/pipecat-ai/pipecat/pull/4983))

- Fixed `ToolsSchemaAdapter` and `LLMContextAdapter` dropping a `ToolsSchema`'s
  `custom_tools` (e.g. Gemini's built-in `google_search` tool) when serializing
  an `LLMContext` for network bus transport, which broke provider-specific
  tools for distributed/multi-worker setups.
  (PR [#4988](https://github.com/pipecat-ai/pipecat/pull/4988))

- Fixed `_send_tool_result()` in the OpenAI, Inworld, and xAI (Grok) realtime
  LLM services double-JSON-encoding tool call results before sending them back
  to the model. Tool results were being serialized twice, so the model would
  receive a JSON string containing an escaped JSON string instead of the actual
  result.
  (PR [#4988](https://github.com/pipecat-ai/pipecat/pull/4988))

- Fixed the Anthropic and AWS Bedrock LLM adapters silently discarding the entire
  conversation when a single message failed to convert to the provider's format
  (e.g. a malformed image data URL). The Anthropic, AWS Bedrock, and Gemini
  adapters now raise `LLMContextConversionError`, surfacing the real cause
  instead of a misleading downstream API error.
  (PR [#4990](https://github.com/pipecat-ai/pipecat/pull/4990))

- Fixed a small amount of audio being dropped from the end of every bot turn.
  `BaseOutputTransport` only flushed complete `audio_out_10ms_chunks`-sized
  chunks to the transport; any trailing audio shorter than one chunk was
  silently discarded when `TTSStoppedFrame` arrived, cutting off the last bit
  of speech (more noticeable with larger `audio_out_10ms_chunks` values). That
  leftover audio is now padded with silence and flushed before the stop frame
  is processed.
  (PR [#4993](https://github.com/pipecat-ai/pipecat/pull/4993))

- Fixed the `multi_worker_handoff` Flows example repeating the assistant's
  reply after handing control back to the router. It now hands off with
  `NO_RESPONSE`, preventing the newly-deactivated worker from responding.
  (PR [#4995](https://github.com/pipecat-ai/pipecat/pull/4995))

- Fixed word-completion tracking (used for bot text sync, interruptions, and
  `text_transforms`) to correctly handle SSML tags in TTS output, such as
  ElevenLabs' `<phoneme alphabet="ipa" ph="...">` tag for custom
  pronunciations. Some TTS providers report a multi-attribute opening tag as
  several separate word-timestamp tokens (e.g. `<phoneme`, `alphabet="ipa"`,
  `ph="...">word`); these fragments are now recognized as markup rather than
  being misrouted or prematurely completing a frame.
  (PR [#5000](https://github.com/pipecat-ai/pipecat/pull/5000))

- Fixed a bug where TTS output containing a lone `<` with no matching `>` (e.g.
  `"5 < 10"` or an emoticon like `"<3"`) got silently truncated in the
  user-facing text and LLM context, because word-completion tracking treated it
  as the start of a truncated SSML tag. Ordinary text like this is now
  preserved in full; only genuine mid-tag word-timestamp fragments (e.g. a
  multi-attribute SSML tag split across several tokens by some TTS providers)
  are still treated as markup.
  (PR [#5002](https://github.com/pipecat-ai/pipecat/pull/5002))

- Fixed Google STT final results waiting for the fallback turn-stop timeout
  instead of finalizing immediately.
  (PR [#5018](https://github.com/pipecat-ai/pipecat/pull/5018))

- Fixed `AnthropicLLMService` and `AWSBedrockLLMService` failing with `400
  invalid_request_error: This model does not support assistant message prefill`
  on Claude 4.6 and newer models. Requests whose message list ends with an
  assistant message — a tool-call preamble committed after tool results, or a
  `filter_incomplete_user_turns` marker landing mid-turn — now automatically
  get a minimal `.` user message appended at request time. The stored
  `LLMContext` is never modified, and models that still support assistant
  prefill (claude-haiku-4-5 and older) are left untouched.
  (PR [#5041](https://github.com/pipecat-ai/pipecat/pull/5041))

- Fixed user turn-stop strategies ending a turn early when an STT service
  finalized a transcript mid-utterance. An interim transcription now clears the
  finalized-transcript fast-path, so a transcript finalized during a pause too
  short for VAD to report a stop no longer skips the STT safety-net timeout
  (or, with a turn analyzer, triggers the turn immediately) while the rest of
  the utterance is still being transcribed.
  (PR [#5043](https://github.com/pipecat-ai/pipecat/pull/5043))

- Fixed traced LLM turns losing their `messages`, `tools`, and system
  instruction span attributes (with an `Error setting up LLM tracing: Object of
  type LLMSpecificMessage is not JSON serializable` warning) when using
  `OpenAIResponsesLLMService` with reasoning enabled. Reasoning messages now
  appear in the traced LLM input with their `encrypted_content` elided.
  (PR [#5047](https://github.com/pipecat-ai/pipecat/pull/5047))

- Fixed TTS spans in OpenTelemetry traces missing some or all of the spoken
  text (regression in 1.2.0): with sentence aggregation, a turn's span only
  kept the last sentence's `text` and `metrics.character_count`, and for
  services that create their audio context inside `run_tts` (e.g. the
  ElevenLabs websocket path) the first sentence was dropped — leaving
  single-sentence turns with no text at all. Text from every `run_tts` call in
  an audio context now accumulates onto the span (joined text, summed character
  count), including text spoken before an interruption.
  (PR [#5049](https://github.com/pipecat-ai/pipecat/pull/5049))

- Fixed the Gemini Live `llm_tool_result` trace span never capturing any tool
    result attributes. The tracing decorator read the decorated `_tool_result`
    method's first positional argument as a dict of result fields, but that
    argument is the `tool_call_id` string, so `tool.call_id`,
    `tool.function_name`, `tool.result`, and `tool.result_status` were silently
    dropped. The decorator now reads the method's positional arguments
    (`tool_call_id`, `tool_call_name`, `result`) directly.
  (PR [#5051](https://github.com/pipecat-ai/pipecat/pull/5051))

- Fixed Gemini Live tracing logging `Error extracting context system
  instructions: 'LLMContext' object has no attribute
  'extract_system_instructions'` on every setup span and never capturing the
  system instruction. `llm_setup` spans now emit the standard
  `gen_ai.system_instructions` attribute, resolved the same way as for other
  LLM services: the service's `settings.system_instruction` takes priority,
  falling back to an initial system message in the context.
  (PR [#5053](https://github.com/pipecat-ai/pipecat/pull/5053))

- Fixed TTS word-timestamp tracking and RTVI progress reporting when using
  `TextAggregationMode.TOKEN`. Previously, streaming tokens directly to the TTS
  service bypassed sentence-level tracking, so word-timestamp services and RTVI
  clients did not receive correct `spoken_status`/progress events. Streamed
  tokens are now regrouped into sentences internally, producing the same
  per-sentence progress frames and word-completion tracking as
  `TextAggregationMode.SENTENCE`.
  (PR [#5066](https://github.com/pipecat-ai/pipecat/pull/5066))

- Fixed `TTSService` applying `append_trailing_space` in
  `TextAggregationMode.TOKEN`: appending a space to every token could split
  words across tokens on services that preserve inter-message whitespace (e.g.
  `RimeTTSService`). The trailing space is now applied in sentence aggregation
  mode only.
  (PR [#5067](https://github.com/pipecat-ai/pipecat/pull/5067))

- Fixed TTS word-level captions dropping or misplacing terminal punctuation for
  languages that put a space before `?` `!` `:` `;` (e.g. French "Comment ça va
  ?"). The punctuation arrives as its own word-timestamp token, which was being
  orphaned when the caption was marked complete on the preceding word —
  dropping the mark from the committed caption, and (for mid-sentence `:` /
  `;`) showing it a word late in progressive captions. The caption now stays
  open until the punctuation token arrives and drains it in place.
  (PR [#5089](https://github.com/pipecat-ai/pipecat/pull/5089))

- Fixed an issue where `XAIHttpTTSService` could intermittently crash the audio
  playback task and leave the bot mute for the rest of the session (`buffer
  size must be a multiple of element size`). xAI's `/v1/tts` endpoint chops its
  PCM stream at arbitrary byte boundaries, so a chunk could end mid-sample; the
  service now emits only whole 16-bit samples.
  (PR [#5090](https://github.com/pipecat-ai/pipecat/pull/5090))

### Other

- Corrected the return type annotation of
  `BaseLLMAdapter.from_standard_tools()` to `list[Any] | NotGiven | None`,
  reflecting that tools passed in as `None` are returned unchanged. Improves
  type-checking accuracy for code built on custom adapters; no runtime behavior
  change.
  (PR [#5078](https://github.com/pipecat-ai/pipecat/pull/5078))

## [1.5.0] - 2026-07-04

### Added

- Added `TogetherSTTService` and `TogetherTTSService` for real-time
  speech-to-text and text-to-speech using Together AI's WebSocket APIs.
  (PR [#4054](https://github.com/pipecat-ai/pipecat/pull/4054))

- Added per-sentence synthesis mode and zero-shot audio prompt support to
  `NvidiaTTSService`, letting NVIDIA TTS users choose between stitched and
  per-request synthesis flows and configure voice-cloning prompts for supported
  models.
  (PR [#4742](https://github.com/pipecat-ai/pipecat/pull/4742))

- Added `on_heartbeat_timeout` event handler to `PipelineWorker`, fired when a
  heartbeat frame is not received within the monitor timeout period.
  (PR [#4761](https://github.com/pipecat-ai/pipecat/pull/4761))

- Added Time To First Audio (TTFA) metrics to TTS services, reported as
  `TTFAMetricsData` alongside the existing TTFB metric. TTFA measures the time
  to the first *audible* sample — TTFB plus the leading silence many providers
  pad onto the start of a response — so comparing the two shows how much
  perceived latency is padding versus service response time. Audible onset is
  detected from short-time RMS energy (`detect_speech_onset` in
  `pipecat.audio.utils`), which rejects noise-floor blips and brief transients;
  the `MetricsLogObserver` surfaces the new metric.
  (PR [#4782](https://github.com/pipecat-ai/pipecat/pull/4782))

- `GeminiTTSService` can now use the Gemini Developer API (google-genai)
  backend in addition to the existing Google Cloud backend. Pass `api_key` (or
  set `GOOGLE_API_KEY`) to authenticate with an API key instead of Google Cloud
  service-account credentials.

    - The backend is selected automatically: passing `api_key` opts into the
  GenAI backend, while `credentials`/`credentials_path` continue to use the
  Google Cloud backend. Use `use_genai=True`/`False` to force a backend
  explicitly. A `GOOGLE_API_KEY` present in the environment alone does not
  switch backends — it is only used once the GenAI backend is active.
    - New `http_options` parameter forwards `google.genai.types.HttpOptions` to
  the GenAI client.
    - The GenAI backend does not support `prompt`/style instructions or
  `multi_speaker` output; setting them logs a warning and they are ignored. Use
  the Google Cloud backend for those features.
  (PR [#4787](https://github.com/pipecat-ai/pipecat/pull/4787))

- Added a `mode` streaming parameter to `AssemblyAISTTService`, exposing
  AssemblyAI's U3 Pro latency/accuracy preset (`min_latency`, `balanced`, or
  `max_accuracy`). It trades transcription accuracy against turn-finalization
  latency and is only applicable to U3 Pro models, where the server defaults to
  `balanced`.
  (PR [#4810](https://github.com/pipecat-ai/pipecat/pull/4810))

- `TaskManager` can now be constructed with an event loop and an optional
  `contextvars.Context` (`TaskManager(loop=..., context=...)`), and creates all
  of its tasks within that context. You can pass a single task manager to
  `WorkerRunner(task_manager=...)` (and to individual workers) to share one
  loop and context across the runner and every worker, so context variables set
  in one task are visible to the others.
  (PR [#4815](https://github.com/pipecat-ai/pipecat/pull/4815))

- Added tunable parameters to the xAI TTS services: `speed`,
  `optimize_streaming_latency`, and `text_normalization` (plus
  `with_timestamps` on the WebSocket service). Set them via the service's
  `Settings`, e.g. `XAITTSService.Settings(speed=1.1)`.
  (PR [#4821](https://github.com/pipecat-ai/pipecat/pull/4821))

- Added word-level timestamps to `XAITTSService`. When `with_timestamps` is
  enabled (now the default), xAI's per-character timing is converted into
  per-word `TTSTextFrame` objects, each carrying an accurate `pts`. Note that
  xAI delivers timestamps in coarse batches, so word frames are emitted in
  bursts; consumers should schedule off `pts` rather than arrival time.
  (PR [#4821](https://github.com/pipecat-ai/pipecat/pull/4821))

- Added a `base_url` parameter to `TwilioFrameSerializer` to configure the REST
  API host used for auto hang-up. By default the host is still derived from
  `region`/`edge` (unchanged behavior), but setting `base_url` lets you target
  a Twilio-API-compatible backend or a self-hosted server instead of
  `api.twilio.com`.
  (PR [#4845](https://github.com/pipecat-ai/pipecat/pull/4845))

- Added a first-class RTVI `dtmf` client message. Sending `{type: "dtmf", data:
  {button: "1"}}` makes the `RTVIProcessor` push an `InputDTMFFrame`
  downstream, the same path a telephony transport's keypress takes, so any bot
  with DTMF handling (e.g. a `DTMFAggregator`) reacts to it. One keypress per
  message.
  (PR [#4849](https://github.com/pipecat-ai/pipecat/pull/4849))

- Added DTMF keypress support to the behavioral evals. A scenario turn can now
  press keys with a `dtmf:` field (e.g. `dtmf: "123#"`) instead of `user:`,
  sent as one RTVI `dtmf` message per key. A bot running a `DTMFAggregator`
  reacts to them as a transcription, so a `dtmf` turn can assert on
  `user_transcription` and `response` like a spoken turn.
  (PR [#4849](https://github.com/pipecat-ai/pipecat/pull/4849))

- Added built-in text transform functions for TTS voice formatting under
  `pipecat.utils.text.transforms`: `strip_markdown`, `normalize_acronyms`,
  `expand_currency`, `expand_numbers`, `expand_percentages`,
  `expand_phone_numbers`, `expand_units`, `email_to_speech`, `normalize_dates`,
  and `replace_text`. These can be composed individually via the
  `text_transforms` parameter on any `TTSService`, or used together via the new
  `VoiceFormatter` bundle.
    - `VoiceFormatter` is a single configurable callable that applies all
  transforms in the correct order (structural cleanup → language expansions →
  custom replacements). Most transforms are enabled by default; pass keyword
  arguments to toggle them:
      ```python
      tts = CartesiaTTSService(
          text_transforms=[("*", VoiceFormatter(expand_numbers=True,
  normalize_acronyms=False))],
      )
      ```
    - Individual transforms can be composed for fine-grained control:
      ```python
      tts = CartesiaTTSService(
          text_transforms=[("*", strip_markdown), ("*", expand_currency), ("*",
  expand_percentages)],
      )
      ```
  (PR [#4854](https://github.com/pipecat-ai/pipecat/pull/4854))

- Added silence-based keepalive to `NvidiaSTTService` to keep idle NVIDIA
  streaming ASR sessions from going stale. When no audio arrives for a while,
  the service sends silence over the existing stream instead of letting it sit
  idle and degrade.
  (PR [#4877](https://github.com/pipecat-ai/pipecat/pull/4877))

- Pipecat Flows is now part of `pipecat-ai`. The conversation-flow framework
  previously published as the separate `pipecat-ai-flows` package now ships
  with Pipecat under the `pipecat.flows` namespace — `from pipecat.flows import
  FlowManager, NodeConfig` — so there is no longer a separate package to
  install or keep version-matched. Code importing from `pipecat_flows` should
  switch to `pipecat.flows`. If the deprecated `pipecat-ai-flows` package is
  still installed alongside this Pipecat, Pipecat logs an error prompting you
  to remove it. The standalone package's release history remains available in
  the archived [pipecat-flows
  repository](https://github.com/pipecat-ai/pipecat-flows/blob/main/CHANGELOG.md).
  (PR [#4882](https://github.com/pipecat-ai/pipecat/pull/4882))

- Added `clear_after_secs` parameter to `SOXRStreamAudioResampler` (default
  `0.2`) to control how long after inactivity the internal resampler state is
  cleared. Set to `None` to disable clearing.
  (PR [#4886](https://github.com/pipecat-ai/pipecat/pull/4886))

- Added `resampler_clear_after_secs` to `FrameSerializer.InputParams` so all
  telephony serializers (Twilio, Plivo, Vonage, Telnyx, Exotel, Genesys) expose
  this setting to callers.
  (PR [#4886](https://github.com/pipecat-ai/pipecat/pull/4886))

- Added a `language_code` streaming parameter to `AssemblyAISTTService` for
  declaring the audio language (e.g. `"es"`, `"fr"`). On U3 Pro models a tier-1
  code (`en`/`es`/`fr`/`de`/`it`/`pt`) steers transcription toward that
  language. It is mutually exclusive with `language_detection` and is not sent
  unless set, so existing behavior is unchanged.
  (PR [#4889](https://github.com/pipecat-ai/pipecat/pull/4889))

- Added `AudioBufferStartRecordingFrame` and `AudioBufferStopRecordingFrame`
  control frames. Push them through the pipeline to start and stop
  `AudioBufferProcessor` recording. The `start_recording()` /
  `stop_recording()` methods continue to work.
  (PR [#4890](https://github.com/pipecat-ai/pipecat/pull/4890))

- Added an `auto_start_recording` option to `AudioBufferProcessor` that starts
  recording as soon as the pipeline starts. Bots generated by the Pipecat CLI
  with the recording feature now use this option.
  (PR [#4890](https://github.com/pipecat-ai/pipecat/pull/4890))

- Added `on_recording_started` and `on_recording_stopped` events to
  `AudioBufferProcessor`, fired when recording starts and stops.
  `on_recording_stopped` fires after the final buffered audio has been emitted.
  (PR [#4890](https://github.com/pipecat-ai/pipecat/pull/4890))

- AI services can now describe themselves to downstream processors at start by
  overriding `service_metadata_frame()` to return a populated
  `ServiceMetadataFrame`; `broadcast_service_metadata()` broadcasts whatever it
  returns. The STT services that do server-side end-of-turn detection (Deepgram
  Flux, Cartesia Turns, AssemblyAI, Gladia, Speechmatics) use this to recommend
  `ExternalUserTurnStrategies`, so bots no longer need to set
  `user_turn_strategies` by hand; your own setting still wins.
  (PR [#4892](https://github.com/pipecat-ai/pipecat/pull/4892))

- Added `endpoint_latency_adjustment_level` to `SonioxSTTService.Settings`,
  exposing Soniox's endpoint-detection latency control (integer 0–3; higher
  finalizes turns sooner at some cost to accuracy). Takes effect when Soniox
  endpoint detection is active (`vad_force_turn_endpoint=False`).
  (PR [#4894](https://github.com/pipecat-ai/pipecat/pull/4894))

- Added a `speed` setting (0.7-1.3) to `SonioxTTSService`.
  (PR [#4947](https://github.com/pipecat-ai/pipecat/pull/4947))

- `SonioxTTSService` now supports cloned voices: pass the voice UUID as
  `voice`.
  (PR [#4947](https://github.com/pipecat-ai/pipecat/pull/4947))

- `SonioxSTTService` now emits `UserStartedSpeakingFrame` /
  `UserStoppedSpeakingFrame` and recommends `ExternalUserTurnStrategies` when
  using Soniox's built-in endpoint detection (`vad_force_turn_endpoint=False`),
  matching `AssemblyAISTTService`. The turn opens on the local VAD signal when
  a VAD analyzer is configured (most responsive) or on the first transcript
  token otherwise, and closes on the Soniox endpoint. A new `should_interrupt`
  parameter (default True) controls whether the bot is interrupted when the
  user starts speaking in this mode.
  (PR [#4949](https://github.com/pipecat-ai/pipecat/pull/4949))

### Changed

- `TavusTransport` now delivers bot audio via the Tavus `conversation.echo` app
  message API instead of a WebRTC audio track. By default, audio is sent paced
  to real playback time (compatible with downstream processors like
  `AudioBufferProcessor`). Set
  `TavusParams(audio_out_faster_than_realtime=True)` to instead accumulate and
  send audio in 100ms chunks as fast as possible, which gives Tavus more of a
  rendering buffer at the cost of losing realtime pacing.
    - The default `persona_id` changed from `"pipecat-stream"` to `"pipecat0"`,
  which signals Tavus to expect audio over the `conversation.echo` app message
  API instead of a WebRTC custom audio track.
  (PR [#4648](https://github.com/pipecat-ai/pipecat/pull/4648))

- `TavusVideoService` now sends bot audio to Tavus via the `conversation.echo`
  app message API instead of a WebRTC audio track. Audio is accumulated and
  sent in 100ms chunks, as fast as possible.
    - The default `persona_id` changed from `"pipecat-stream"` to `"pipecat0"`,
  which signals Tavus to expect audio over the `conversation.echo` app message
  API instead of a WebRTC custom audio track.
  (PR [#4648](https://github.com/pipecat-ai/pipecat/pull/4648))

- ⚠️ The default `GeminiTTSService` model changed from `gemini-2.5-flash-tts`
  to `gemini-3.1-flash-tts-preview`. Pass
  `settings=GeminiTTSService.Settings(model="gemini-2.5-flash-tts")` to keep
  the previous model.
  (PR [#4787](https://github.com/pipecat-ai/pipecat/pull/4787))

- `TTFAMetricsData` now reports the latency breakdown directly: `ttfa` (the
  measurement, renamed from `value`), `ttfb`, and `leading_silence`. Consumers
  can see how much of the perceived latency is silence padding
  (`leading_silence == ttfa - ttfb`) without correlating a separate
  `TTFBMetricsData`. `ttfb` here mirrors the standalone, earlier
  `TTFBMetricsData` for convenience and is not a separate measurement.
  (PR [#4814](https://github.com/pipecat-ai/pipecat/pull/4814))

- Dangling tasks are now reported by the `WorkerRunner` for its shared task
  manager once everything has been torn down. A worker only reports dangling
  tasks when it owns its own task manager, so a worker sharing the runner's
  task manager no longer flags the runner's and other workers' tasks as
  dangling. `check_dangling_tasks` is now a constructor argument on both
  `WorkerRunner` and `BaseWorker`.
  (PR [#4815](https://github.com/pipecat-ai/pipecat/pull/4815))

- `XAITTSService` now cancels the current utterance on interruption by sending
  a `text.clear` message over the existing WebSocket, instead of disconnecting
  and reconnecting. This makes barge-in faster by avoiding a reconnect on every
  interruption.
  (PR [#4821](https://github.com/pipecat-ai/pipecat/pull/4821))

- Bumped `pipecat-ai-prebuilt` minimum version to `1.0.3` in the `runner`
  extra, which updates the prebuilt client UI served by the development runner
  to use RTVI protocol 2.0.0.
  (PR [#4847](https://github.com/pipecat-ai/pipecat/pull/4847))

- Eval scenarios now accept a `send_after:` with only `delay_ms` (no `event`),
  a pure time delay relative to the previous send, and `expect:` is now
  optional so a turn can just send input or wait.
  (PR [#4849](https://github.com/pipecat-ai/pipecat/pull/4849))

- Simplified the `WorkerBus` message-dispatch tasks by removing redundant
  `CancelledError` handling (the task manager already handles cancellation, and
  `CancelledError` was never caught by the subscriber-isolation handler).
  Subscriber-exception isolation and cancellation behavior are unchanged.
  (PR [#4851](https://github.com/pipecat-ai/pipecat/pull/4851))

- When text transforms change the alphanumeric content of a TTS frame (e.g.
  `expand_currency` turning `"$42.50"` into `"forty-two dollars and fifty
  cents"`), the conversation context now correctly receives the original LLM
  text (`"$42.50"`) rather than the expanded TTS words. Intermediate spoken
  words within a transformed span are suppressed from the context until the
  full span is complete, so the context entry is clean and accurately
  represents what was said.
  (PR [#4854](https://github.com/pipecat-ai/pipecat/pull/4854))

- `pipecat init` is now the starting point for building a Pipecat app. It
  writes the coding-agent files `AGENTS.md` and `CLAUDE.md`, then helps you
  build:
    - Build with a coding agent (such as Claude Code or Codex). This also
  writes a `GETTING_STARTED.md` guide for building Pipecat apps with an AI
  coding assistant.
    - Scaffold a runnable bot immediately through an interactive setup wizard.
    - Run `pipecat init quickstart` to scaffold the ready-to-run quickstart
  project, set up for coding agents in one step.
  (PR [#4861](https://github.com/pipecat-ai/pipecat/pull/4861))

- `AssemblyAISTTService` now defaults to the `universal-3-5-pro` model
  (AssemblyAI's launched flagship streaming model), replacing the pre-GA alias
  `u3-rt-pro`. Both resolve to the same U3 Pro family and feature set, so the
  only change is the model id sent on the wire. `u3-rt-pro` and
  `u3-rt-pro-beta-1` remain accepted for backward compatibility.
  (PR [#4863](https://github.com/pipecat-ai/pipecat/pull/4863))

- Bumped the `@pipecat-ai/*` JS client dependencies used by the `pipecat init`
  client templates (and the UI-worker examples): `client-js` to `1.12.0`,
  `client-react` to `1.7.1`, `daily-transport` to `1.6.7`,
  `small-webrtc-transport` to `1.10.5`, and `websocket-transport` to `1.7.0`.
  (PR [#4866](https://github.com/pipecat-ai/pipecat/pull/4866))

- ⚠️ `pipecat init` now keeps existing `AGENTS.md`, `CLAUDE.md`, and
  `GETTING_STARTED.md` files instead of overwriting them on re-run. When a
  guide was written by an older Pipecat version, an interactive run offers to
  refresh it; otherwise pass the new `--overwrite-guide` flag (renamed from
  `--force`, now covering all three files) to refresh them.
  (PR [#4869](https://github.com/pipecat-ai/pipecat/pull/4869))

- Updated the ai-coustics integration to SDK 0.21 (bumped `aic-sdk` to
  `~=2.5.0`). `AICQuailVADAnalyzer` now reports the model's continuous raw VAD
  probability (`VadContext.raw_vad_probability()`) gated by Pipecat's
  `VADParams` instead of a binary speech flag. Because the previous output was
  binary (`0.0`/`1.0`), `VADParams.confidence` had no effect on this analyzer
  before — it now governs the speech threshold, so existing
  `AICQuailVADAnalyzer` users should review their `VADParams.confidence` after
  upgrading. The ai-coustics voice examples now use the `quail-vf-2.2-l-16khz`
  enhancement model.
  (PR [#4874](https://github.com/pipecat-ai/pipecat/pull/4874))

- Widened the `google` extra's `google-genai` dependency to `>=1.68.0,<3`,
  allowing the 2.x line. The google-genai 2.0 major release scopes its breaking
  changes to the Interactions API, which Pipecat does not use; the surfaces
  Pipecat relies on (`Client`, `aio.live.connect`,
  `generate_content`/`generate_content_stream`, `types.*`) are unchanged, so no
  migration is required.
  (PR [#4895](https://github.com/pipecat-ai/pipecat/pull/4895))

- ⚠️ `LLMContextAggregatorPair`'s `realtime_service_mode` is now
  auto-configured and defaults to `None` (was `False`): a realtime
  (speech-to-speech) LLM service announces itself via service metadata and the
  aggregator turns the mode on automatically, so you no longer opt in by hand.
  If your realtime-service pipeline previously ran without
  `realtime_service_mode=True`, realtime context-write behavior now applies to
  it: listen for `on_user_turn_message_added` to get the newly-added user
  message rather than `on_user_turn_stopped`, which no longer carries it
  (`UserTurnStoppedMessage.content` is `None` in realtime mode). Pass
  `realtime_service_mode=False` to keep the legacy, pre-`realtime_service_mode`
  behavior.
  (PR [#4919](https://github.com/pipecat-ai/pipecat/pull/4919))

- ⚠️ Auto-switching to external user turn strategies for realtime services is
  no longer conditioned on `realtime_service_mode`: a realtime service that
  does its own server-side turn detection now gets `ExternalUserTurnStrategies`
  whenever it recommends them, even with `realtime_service_mode=False` or left
  off. (The switch has moved onto the same service-metadata recommendation
  mechanism the STT services use.) As before, passing your own
  `user_turn_strategies` overrides the recommendation.
  (PR [#4919](https://github.com/pipecat-ai/pipecat/pull/4919))

- ⚠️ `SonioxTTSService` now emits word-aligned `TTSTextFrame`s instead of
  pushing each sentence's full text up front, so the context reflects only what
  was actually spoken on interruption.
  (PR [#4947](https://github.com/pipecat-ai/pipecat/pull/4947))

### Deprecated

- Deprecated `TaskManagerParams` and `TaskManager.setup()`. Pass `loop` and
  `context` to the `TaskManager` constructor instead.
  - Deprecated the `WorkerRunner` `loop` argument. Pass `task_manager` (which
  owns its own loop) instead.
  (PR [#4815](https://github.com/pipecat-ai/pipecat/pull/4815))

- Deprecated the `speech_hold_duration`, `minimum_speech_duration`, and
  `sensitivity` parameters of `AICQuailVADAnalyzer`. They only affected the
  SDK's post-processed VAD output, which the new raw-probability path no longer
  uses — speech gating is now governed by Pipecat's `VADParams`
  (`confidence`/`start_secs`/`stop_secs`). The parameters are accepted but
  ignored, and will be removed in 2.0.0.
  (PR [#4874](https://github.com/pipecat-ai/pipecat/pull/4874))

### Removed

- ⚠️ Removed `WorkerParams.loop`. Pass a task manager via
  `WorkerParams(task_manager=...)` instead.
  (PR [#4815](https://github.com/pipecat-ai/pipecat/pull/4815))

- ⚠️ Removed the `pipecat create` command; scaffolding now lives in `pipecat
  init`, the single entry point for starting a Pipecat app. Alongside the
  coding-agent guide (`AGENTS.md`, `CLAUDE.md`) it already wrote, `pipecat
  init` now also scaffolds a runnable bot — interactively, or non-interactively
  from flags or a config file (e.g. `pipecat init . --bot-type web -t daily
  --stt deepgram_stt --llm openai_llm --tts cartesia_tts`; run `pipecat init
  --list-options` for valid values). `pipecat init quickstart` replaces
  `pipecat create quickstart`. Scaffolding is now directory-first and in-place
  — the project name comes from the target directory, and `create`'s
  `--output/-o` and `--name`-subfolder layout are gone.
  (PR [#4883](https://github.com/pipecat-ai/pipecat/pull/4883))

- ⚠️ Removed the internal `RealtimeServiceMetadataFrame` and
  `RealtimeServiceInfo`. Realtime LLM services now describe themselves with
  `LLMServiceMetadataFrame` (carrying `is_realtime_service`); if you imported
  either symbol directly, switch to `LLMServiceMetadataFrame`.
  (PR [#4919](https://github.com/pipecat-ai/pipecat/pull/4919))

### Fixed

- Fixed `MarkdownTextFilter` leaking raw `#` markers into TTS output for
  second-level (and deeper) markdown headers. Headers are now normalized to
  plain text before the newline-collapse step that previously broke header
  recognition by `md.convert`. This also handles closed ATX headers (e.g. `##
  Title ##`) while preserving trailing whitespace needed for word-by-word
  streaming.
  (PR [#4708](https://github.com/pipecat-ai/pipecat/pull/4708))

- Fixed `NvidiaSegmentedSTTService` not initializing `speaker_diarization` and
  `diarization_max_speakers` defaults, which left both fields as `NOT_GIVEN`
  after construction.
  (PR [#4718](https://github.com/pipecat-ai/pipecat/pull/4718))

- Fixed `SarvamTTSService` WebSocket handshakes sending duplicate `User-Agent`
  headers. Pipecat now passes the Sarvam SDK `User-Agent` via the WebSocket
  client's `user_agent_header` parameter instead of `additional_headers`.
  (PR [#4794](https://github.com/pipecat-ai/pipecat/pull/4794))

- Fixed `DeepgramSageMakerSTTService` blocking pipeline startup while
  connecting. The SageMaker BiDi connection is now established in a background
  task, so a slow or failing connect no longer holds up the `StartFrame`
  barrier (the first bot turn, e.g. a greeting, can proceed while STT connects)
  and connection failures surface via `on_connection_error` instead of looking
  like a hang.
  (PR [#4803](https://github.com/pipecat-ai/pipecat/pull/4803))

- Fixed `RNNoiseFilter` failing to import with `No module named 'av.option'`
  when installing `pipecat-ai[rnnoise]`. PyAV 17.1.0 removed the `av.option`
  submodule that `pyrnnoise`'s `audiolab` dependency imports, so the `rnnoise`
  extra now caps `av<17.1.0`.
  (PR [#4807](https://github.com/pipecat-ai/pipecat/pull/4807))

- Fixed eval failure reasons (a missing function call, a response timeout, an
  unsatisfied `eval:`, or a `send_after` that never fired) not being written to
  the per-scenario `.eval.log` file. They were only printed to the terminal
  during the run, so there was no record to debug failures after the fact.
  (PR [#4811](https://github.com/pipecat-ai/pipecat/pull/4811))

- Fixed `SileroOnnxModel` raising a confusing `AttributeError` instead of a
  `ValueError` when given an input audio chunk with too many dimensions. The
  validation error message called `x.dim()` (a PyTorch method) on a NumPy
  array; it now uses `x.ndim` and surfaces the intended "Too many dimensions"
  message.
  (PR [#4820](https://github.com/pipecat-ai/pipecat/pull/4820))

- Fixed `XAIHttpTTSService` omitting the `language` field when it was unset.
  xAI marks `language` as required, so it is now always sent, falling back to
  `"auto"` for language auto-detection.
  (PR [#4821](https://github.com/pipecat-ai/pipecat/pull/4821))

- Fixed `WorkerBus` permanently stopping message delivery to a subscriber when
  that subscriber's `on_bus_message` (or an overridable lifecycle hook such as
  `on_job_response`/`on_activated`) raised an exception. The router and data
  dispatch tasks now log the exception and keep running, so subsequent messages
  — including cancel/cleanup — are still delivered.
  (PR [#4827](https://github.com/pipecat-ai/pipecat/pull/4827))

- Fixed `Mem0MemoryService` injecting an empty "Based on previous
  conversations, I recall:" message into the LLM context on turns where no
  relevant memories were found. Mem0 2.x `search()` returns a `{"results":
  [...]}` dict, which is always truthy, so the empty-memory guard never
  triggered; retrieved memories are now normalized to a list so the guard,
  formatting, and debug counts all behave correctly.
  (PR [#4843](https://github.com/pipecat-ai/pipecat/pull/4843))

- Fixed a missing `pipecat/workers/__init__.py` so the `py.typed` marker covers
  `pipecat.workers.*` and type checkers (e.g. pyright in strict mode) resolve
  `from pipecat.workers.runner import WorkerRunner` without reporting
  `reportMissingTypeStubs`.
  (PR [#4846](https://github.com/pipecat-ai/pipecat/pull/4846))

- Fixed `RTVIObserver` silently dropping TTFA (Time To First Audio) metrics.
  TTFA metrics are now forwarded to RTVI clients under a `ttfa` key, alongside
  TTFB, processing, token, and character metrics.
  (PR [#4880](https://github.com/pipecat-ai/pipecat/pull/4880))

- Fixed `TwilioFrameSerializer` rejecting a valid `base_url` configuration when
  only one of `region`/`edge` was also set. The `region`/`edge` pairing is only
  required when deriving Twilio's FQDN host; since `base_url` is used verbatim
  and ignores `region`/`edge`, the validation now skips that check when
  `base_url` is provided.
  (PR [#4885](https://github.com/pipecat-ai/pipecat/pull/4885))

- Fixed eval scenarios silently corrupting unquoted DTMF turns with leading
  zeros or a hex prefix. YAML 1.1 parsed `dtmf: 012` as octal (`10`) and `dtmf:
  0x10` as hex before validation, sending the wrong keypresses; the scenario
  loader now resolves only plain-decimal scalars as integers, so leading-zero
  sequences keep their digits and `dtmf: 123` still works.
  (PR [#4887](https://github.com/pipecat-ai/pipecat/pull/4887))

- Fixed local STT services (`WhisperSTTService`, `WhisperSTTServiceMLX`,
  `MoonshineSTTService`) corrupting the start of every utterance.
  `SegmentedSTTService` wraps each VAD segment in a WAV container, but these
  services read the bytes directly as 16-bit PCM, so the 44-byte WAV header was
  decoded as 22 `int16` samples and prepended as a near-full-scale burst,
  changing the transcription. `SegmentedSTTService` now exposes a
  `wants_wav_segments` property (default `True`, what cloud upload APIs expect)
  that local models override to receive raw PCM instead.
  (PR [#4896](https://github.com/pipecat-ai/pipecat/pull/4896))

- Fixed services and transports leaking connections and background tasks when a
  pipeline is torn down without an `EndFrame`/`CancelFrame` reaching every
  processor. Resource release (closing websockets and connections, releasing
  clients/sessions, cancelling `create_task()` tasks) now runs from the
  guaranteed `cleanup()` hook in addition to the frame-driven
  `stop()`/`cancel()` paths, so it happens on every exit path. Affected
  services include the websocket STT/TTS bases (and their subclasses), realtime
  LLM services, Deepgram Flux, and the
  AWS/Azure/Google/NVIDIA/HeyGen/Simli/Tavus integrations, plus the Daily,
  LiveKit, websocket, SmallWebRTC, and Tavus transports.
  (PR [#4902](https://github.com/pipecat-ai/pipecat/pull/4902))

- Fixed occasional abnormal WebSocket closures (1006) when disconnecting the
  ElevenLabs TTS service. Pipecat now waits for ElevenLabs to complete its side
  of the two-step close before closing, rather than racing the closing
  handshake.
  (PR [#4904](https://github.com/pipecat-ai/pipecat/pull/4904))

- Fixed `InworldTTSService` surfacing an `ErrorFrame` during idle periods when
  the keepalive task sent a contextless `send_text` and Inworld rejected it.
  The `context_id is required` and `no open context` rejections are now treated
  as benign (logged at debug level and skipped), matching the existing `Context
  not found` handling, which prevents spurious failover away from Inworld.
  (PR [#4906](https://github.com/pipecat-ai/pipecat/pull/4906))

- Fixed `ElevenLabsHttpTTSService` sending `previous_text` with the `eleven_v3`
  model, which rejects that parameter. ElevenLabs returned a 400 error for
  every request after the first sentence of a turn, so only the first sentence
  of a multi-sentence response was spoken.
  (PR [#4925](https://github.com/pipecat-ai/pipecat/pull/4925))

- Fixed `AggregatedFrameSequencer` duplicating a word and misattributing its
  context in TTS word-timestamp mode (Cartesia, ElevenLabs, etc.) when a
  whitespace-only token was force-completed.
  (PR [#4930](https://github.com/pipecat-ai/pipecat/pull/4930))

- Fixed the incomplete-turn (`○`/`◐`) re-prompt nudge firing while the user was
  already speaking again. The re-prompt timeout was only cancelled on
  `InterruptionFrame`, which does not fire when the user resumes speaking
  inside the same open turn. It is now also cancelled on
  `VADUserStartedSpeakingFrame`, so a user who resumes after a pause is no
  longer interrupted by a canned "no rush" prompt.
  (PR [#4938](https://github.com/pipecat-ai/pipecat/pull/4938))

- Fixed the bot speaking the same response multiple times within one user turn
  when using `filter_incomplete_user_turns` turn completion. When the acoustic
  detector (e.g. Smart Turn) triggered several inferences in one turn, each
  produced a `✓` and every one was voiced. At most one completion is now spoken
  at a time; later duplicate completions are dropped until a new user turn
  begins or the user resumes speaking within the same turn.
  (PR [#4938](https://github.com/pipecat-ai/pipecat/pull/4938))

- Fixed a second, redundant LLM inference when using
  `filter_incomplete_user_turns` turn completion. After the LLM marked a user
  turn incomplete (○/◐), the mixin armed a re-prompt timeout; if that timeout
  fired at the same moment a completed inference (✓) arrived, both the
  re-prompt and the completed response ran. The pending timeout is now
  cancelled as soon as a new LLM response starts, so only one inference runs.
  (PR [#4938](https://github.com/pipecat-ai/pipecat/pull/4938))

- Fixed the bot talking over the user when using `filter_incomplete_user_turns`
  turn completion. A completion (`✓`) resolves with some latency, so the user
  may have resumed speaking by the time it arrives. The user turn controller
  now drops a turn finalization that arrives while the user is speaking, so a
  stale completion no longer ends the turn (and talks over them); the turn
  stays open for the next inference to re-evaluate.
  (PR [#4938](https://github.com/pipecat-ai/pipecat/pull/4938))

- Fixed `AssemblyAISTTService` processing metrics never being recorded in
  AssemblyAI turn-detection mode (`vad_force_turn_endpoint=False`) with
  `should_interrupt=True` (the default): the interruption broadcast on speech
  start immediately stopped the just-started metrics. Metrics now start after
  the interruption broadcast.
  (PR [#4949](https://github.com/pipecat-ai/pipecat/pull/4949))

- Fixed `RimeTTSService` intermittently going silent for the rest of the
  session when metrics are enabled. Rime's websocket may split a 16-bit sample
  across audio chunks, and the resulting odd-length audio frames crashed the
  TTS playback task; dangling bytes are now carried over to the next chunk so
  frames always contain whole samples.
  (PR [#4952](https://github.com/pipecat-ai/pipecat/pull/4952))

## [1.4.0] - 2026-06-16

### Added

- Added `on_user_turn_message_added` event handler on `LLMUserAggregator`, with
  a new `UserTurnMessageAddedMessage` arg type. It fires when the user
  aggregator writes a message to the LLM context, carrying the finalized turn
  text. In cascade mode it coincides with `on_user_turn_stopped`; in realtime
  mode (when `realtime_service_mode=True` on the aggregator pair) it's the
  canonical way to subscribe to "context just updated, here's the user text"
  (since the `on_user_turn_stopped` event fires before the message is
  finalized, with `UserTurnStoppedMessage.content=None`). Note that there's
  been no change to `on_assistant_turn_stopped`.
  (PR [#4533](https://github.com/pipecat-ai/pipecat/pull/4533))

- Added `RealtimeServiceMetadataFrame`, broadcast at pipeline start by realtime
  LLM services (OpenAI Realtime, Azure Realtime, Inworld, Grok/xAI Realtime,
  Gemini Live, AWS Nova Sonic, Ultravox). This frame can be used by other
  processors in the pipeline to configure themselves accordingly. Today, it
  only advertises two things: that a realtime service is present in the
  pipeline (indicated by the fact that the frame is sent at all), and
  `emits_user_turn_frames`, which says whether the realtime service can emit
  its own `UserStartedSpeakingFrame` and `UserStoppedSpeakingFrame`s
  (suggesting local VAD/turn detection may not be needed in the pipeline).
  (PR [#4533](https://github.com/pipecat-ai/pipecat/pull/4533))

- Added to our examples "locally-driven-turns" variants for:
    - OpenAI Realtime (`realtime-openai-locally-driven-turns.py`)
    - Grok Realtime (`realtime-grok-locally-driven-turns.py`)
    - Inworld Realtime (`realtime-inworld-locally-driven-turns.py`)

  These join `realtime-gemini-live-locally-driven-turns.py` in showing how to
  configure each realtime service so that its turn-taking is dictated by local
  turn detection (e.g. VAD + smart turn analyzer).
  (PR [#4533](https://github.com/pipecat-ai/pipecat/pull/4533))

- Added a startup WARNING log on realtime LLM services that don't emit
  `UserStartedSpeakingFrame`/`UserStoppedSpeakingFrame` (Gemini Live, AWS Nova
  Sonic, Ultravox). The log is meant to draw attention to a couple of things:
    - That other processors in the pipeline (e.g. RTVI) may expect turn frames,
      and that the developer can enable local VAD/turn detection to supply them,
      and, relatedly
    - That when using local turn detection, local turns may NOT perfectly align
      with the "ground truth" of server-decided turns, so they should be thought of
      as APPROXIMATE (unless local turn detection is _driving_ the realtime
  service's turns, in which case there's no separate server-decided ground
  truth)

  (The warning also serves as a little nudge to the realtime service
  providers: providing a "ground truth" signal of when the provider thinks the
  user has started or stopped speaking is very helpful to app developers!)
  (PR [#4533](https://github.com/pipecat-ai/pipecat/pull/4533))

- Added a `realtime_service_mode: bool` kwarg on `LLMContextAggregatorPair`,
  for opting into a set of behaviors tailored for use with realtime
  (speech-to-speech) services. Setting `realtime_service_mode=True` does three
  things:
    1. _Decouples context writes from the `UserStoppedSpeakingFrame` signal._
       Instead, the assistant response start triggers the user message
       writes. This ensures that context is written properly even when the
       realtime service provides no turn frames and local turn detection
       (i.e. local VAD) is disabled. This mechanism also enables the next point.
    2. _Lets `UserStoppedSpeakingFrame` fire without waiting for transcripts._
       When local turn detection is configured to drive realtime service
       conversations, `UserStoppedSpeakingFrame` is the signal that triggers
       assistant responses. By letting this frame fire earlier, we reduce
       latency.
    3. _Replaces the default turn strategies with
       `ExternalUserTurnStartStrategy` and `ExternalUserTurnStopStrategy` when
       the realtime service advertises that it emits its own turn frames._
       Various realtime services (OpenAI Realtime, Azure, Grok, Inworld) emit
       their own turn frames; in that case the External strategies fire
       `on_user_turn_started` / `on_user_turn_stopped` from the server-emitted
       `UserStartedSpeakingFrame` / `UserStoppedSpeakingFrame`. For realtime
       services that don't emit those frames — either because they never do
       (Gemini Live, Nova Sonic, Ultravox) or because server-side turn detection
       has been disabled at runtime (e.g. OpenAI Realtime with
       `turn_detection=False`, in locally-driven-turns setups) — the defaults
       stay in place so locally-driven turn detection (e.g. local VAD) can fire
       the events. Passing custom `user_turn_strategies` opts out of the swap.

  Note that when `realtime_service_mode=True`, you should listen for the new
  `on_user_turn_message_added` event to get the newly-added user message rather
  than `on_user_turn_stopped`, which no longer carries it.
  (PR [#4533](https://github.com/pipecat-ai/pipecat/pull/4533))

- Added `private_endpoint` parameter to `AzureTTSService` and
  `AzureHttpTTSService` for connecting via Private Link or custom domain
  endpoints, matching existing `AzureSTTService` support.
  (PR [#4549](https://github.com/pipecat-ai/pipecat/pull/4549))

- Added `will_be_spoken` field to `AggregatedTextFrame`. Set to `True` by the
  TTS service just before synthesis, allowing downstream processors and
  observers to know whether TTS will speak a given text segment before audio
  begins.
  (PR [#4559](https://github.com/pipecat-ai/pipecat/pull/4559))

- Added `AggregatedTextProgressFrame` — a new frame emitted alongside each
  `TTSTextFrame` during word-timestamp playback. It carries `accumulated_text`
  (text already spoken) and `remaining_text` (text not yet spoken) for the
  active segment, enabling downstream consumers such as the RTVI observer to do
  word-level highlighting without coupling to internal sequencer state.
  (PR [#4559](https://github.com/pipecat-ai/pipecat/pull/4559))

- Added `AICQuailVADAnalyzer` (`pipecat.audio.vad.aic_quail_vad`), a
  noise-robustVoice Activity Detection analyzer powered by the standalone Quail
  VAD 2.0 model from the ai-coustics SDK (`aic-sdk~=2.3.0`). It owns its own
  `Processor` and works independently of `AICFilter`, so it can sit before or
  after enhancement in the pipeline. Defaults to the published
  `quail-vad-2.0-xxs-16khz` model; supply `model_id`/`model_path` to override.
  (PR [#4588](https://github.com/pipecat-ai/pipecat/pull/4588))

- Added `continuous_partials` and `interruption_delay` connection parameters to
  the AssemblyAI streaming STT service (`u3-rt-pro` only).
  `continuous_partials` defaults to `True` so voice agents receive interim
  transcripts at a steady cadence during long turns; `interruption_delay`
  (0–1000 ms) overrides how soon the first partial is emitted. Both are exposed
  via `AssemblyAISTTService.Settings` and are omitted for non-`u3-rt-pro`
  models.
  (PR [#4593](https://github.com/pipecat-ai/pipecat/pull/4593))

- Added a `user_audio_preroll_secs` parameter to `GeminiLiveLLMService`
  controlling how much "pre-roll" audio is replayed (sent to Gemini Live) when
  the user turn start is confirmed, in locally-driven-turns mode (server-side
  VAD disabled). Defaults to `None`, auto-sizing the pre-roll duration from the
  upstream VAD's `start_secs` (which assumes VAD drives turn starts); set it
  explicitly when using a non-VAD turn-start strategy.
  (PR [#4597](https://github.com/pipecat-ai/pipecat/pull/4597))

- Added a `user_audio_preroll_secs` parameter to `OpenAIRealtimeLLMService`
  controlling how much "pre-roll" audio is replayed (re-appended to the input
  audio buffer) when the user turn start is confirmed, in locally-driven-turns
  mode (server-side turn detection disabled). Defaults to `None`, auto-sizing
  the pre-roll duration from the upstream VAD's `start_secs` (which assumes VAD
  drives turn starts); set it explicitly when using a non-VAD turn-start
  strategy.
  (PR [#4599](https://github.com/pipecat-ai/pipecat/pull/4599))

- Added word-level timestamp support to `SmallestTTSService`. Enabled by default
  via the `word_timestamps` constructor argument, it emits per-word
  `TTSTextFrame`s aligned to audio playback so downstream consumers (captions,
  lip-sync, RTVI) receive word timing. Timestamps from each TTS request are
  offset onto the turn's continuous playback timeline, so multi-sentence turns
  stay correctly ordered. Available on Smallest's word-timestamp-capable voices;
  other voices simply emit no word events, so leaving it on is safe. Pass
  `word_timestamps=False` to fall back to whole-text frames.
  (PR [#4612](https://github.com/pipecat-ai/pipecat/pull/4612))

- Added a `profanity` setting to `AzureSTTService` (via
  `settings=AzureSTTService.Settings(profanity=...)`) controlling how Azure
  handles profanity in transcripts. Accepts `"raw"` (no masking), `"masked"`
  (Azure default, replaces profane words with `****`), or `"removed"` (drops
  profane words). Defaults to `None` (keeps the Azure SDK default of
  `"masked"`). Use `"raw"` for non-English deployments where Azure's profanity
  list over-eagerly masks ordinary words. The setting is runtime-updatable and
  triggers a reconnect when changed.
  (PR [#4620](https://github.com/pipecat-ai/pipecat/pull/4620))

- WhatsApp `connection_callback` now receives the full call metadata
  (`WhatsAppConnectCall`) as a second argument, available in bot code via
  `runner_args.body`. This gives bots access to the caller's phone number, call
  ID, direction, and timestamp without any extra API calls.
  (PR [#4622](https://github.com/pipecat-ai/pipecat/pull/4622))

  Added the `pipecat create` project-scaffolding CLI to `pipecat-ai`, available
  via the optional `cli` extra. Install it with `uv tool install
  "pipecat-ai[cli]"` (add `--with pipecatcloud` to enable `pipecat cloud`),
  then run `pipecat create` to scaffold a new bot project. The CLI dependencies
  are optional, so they are not pulled into a plain `pip install pipecat-ai`.
  (PR [#4631](https://github.com/pipecat-ai/pipecat/pull/4631))

  `pipecat create` takes an optional target directory: pass a path — for example
  `pipecat create .` — to scaffold **directly into that directory** (the same
  convention as `npm create vite@latest .`), or omit it to nest the project
  under a `<project-name>/` subfolder. The project name defaults to the target
  directory's basename, and `--name` overrides it.
  (PR [#4631](https://github.com/pipecat-ai/pipecat/pull/4631))

- Added `websocket` to the development runner's `-t`/`--transport` choices, so
  you can now run `python bot.py -t websocket` to restrict the server to the
  plain WebSocket transport (served at `/ws-client`). The startup banner prints
  a websocket-specific message with the prebuilt UI and
  `ws(s)://host:port/ws-client` endpoint.
  (PR [#4636](https://github.com/pipecat-ai/pipecat/pull/4636))

- Direct functions advertised in an `LLMContext` are now registered
  automatically — no separate registration call. List a direct function in
  `LLMContext(tools=[...])`, or push an `LLMSetToolsFrame` to change tools
  mid-session, and its handler is registered. The advertised tool set is the
  single source of truth: dropping a direct function unregisters its handler
  too. Also applies across `LLMSwitcher` member LLMs.
  (PR [#4654](https://github.com/pipecat-ai/pipecat/pull/4654))

- `LLMContext(tools=...)` and `LLMSetToolsFrame` now accept a plain list of
  direct functions and/or `FunctionSchema` objects, not just a `ToolsSchema`.
  (PR [#4654](https://github.com/pipecat-ai/pipecat/pull/4654))

- Added an optional `@tool_options(cancel_on_interruption=...,
  timeout_secs=...)` decorator for overriding a direct function's call options;
  defaults apply otherwise.
  (PR [#4654](https://github.com/pipecat-ai/pipecat/pull/4654))

- Added `PipelineFlushFrame`, a control frame for draining the pipeline. Push
  it downstream and the pipeline worker bounces it back upstream so it
  round-trips through every processor, then sets its `event`. Await that event
  to know all in-flight frames queued ahead of the probe have been processed
  (e.g. to let the pipeline settle after an interruption before injecting a new
  frame). It's an `UninterruptibleFrame`, so the probe survives an
  `InterruptionFrame` and still completes its round-trip.
  (PR [#4655](https://github.com/pipecat-ai/pipecat/pull/4655))

- Added `pipecat.evals`, a behavioral eval framework for Pipecat bots, usable
  both as a library and from the CLI. A YAML scenario describes a scripted
  conversation and the semantic events expected back from the bot
  (transcriptions, LLM/TTS responses, function calls) with optional latency
  budgets and natural-language criteria judged by an LLM, in text or audio mode
  (audio synthesizes the user's speech and transcribes the bot's actual audio).
  In code, `EvalScenario.load()` parses a scenario and
  `EvalSession.from_scenario(...).run()` runs it against a bot, returning a
  structured `EvalResult` (with `EvalManifest.load()` and `EvalSuite.run()` for
  the multi-bot path). The new `pipecat eval run` (against an already-running
  bot) and `pipecat eval suite` (a manifest mapping bots to the scenarios they
  run) commands wrap the same library and are also reachable as `python -m
  pipecat.evals`. Bots opt in by exposing the `-t eval` transport.
  (PR [#4655](https://github.com/pipecat-ai/pipecat/pull/4655))

- Added a `bot-interrupted` RTVI server message, emitted when the bot's
  in-flight output is cut off (a VAD-detected user barge-in or a programmatic
  interrupt), so clients can drop whatever the bot was mid-saying.
  (PR [#4655](https://github.com/pipecat-ai/pipecat/pull/4655))

- Added an opt-in `--eval` flag to `pipecat create` (and an `Enable evals?`
  wizard prompt, off by default) that makes the generated bot eval-ready
  without any manual edit:
    - an `"eval"` entry in the bot's `transport_params`, so the bot is runnable
      with `-t eval`. The entry mirrors the bot's audio/video settings and is inert
      unless the bot is run with `-t eval`.
    - runnable starter scenarios in `server/evals/` that pass against the
      freshly scaffolded bot and double as schema references to copy when adding
      more: `starter_text.yaml` (text mode, the fast inner loop; cascade bots only)
      and `starter_audio.yaml` (the full audio round trip, the only mode for
      realtime speech-to-speech bots).
    - the dependencies to run them from the project's own environment: the
      `cli` extra (the `pipecat eval` command) plus `kokoro` and `moonshine` (the
      harness's local speech stack), so audio-mode evals run with no extra setup
      and no API keys.
  (PR [#4664](https://github.com/pipecat-ai/pipecat/pull/4664))

- Added `filter_repeated_sequences` parameter to
  `MarkdownTextFilter.InputParams` to allow disabling repeated sequence
  removal.
  (PR [#4674](https://github.com/pipecat-ai/pipecat/pull/4674))

- Added support for Belgium german in transcription languages
  (PR [#4682](https://github.com/pipecat-ai/pipecat/pull/4682))

- Added `MoonshineSTTService`, a local speech-to-text service backed by
  [Moonshine](https://github.com/moonshine-ai/moonshine). It runs a small, fast
  ASR model on the CPU via ONNX Runtime, so it needs no GPU and no API key (the
  model downloads once on first use and is cached). Install with `pip install
  "pipecat-ai[moonshine]"` and choose the model via
  `MoonshineSTTService.Settings(model=...)` (a `Model` enum member or string):
  `Model.TINY`, `Model.BASE`, or a streaming model run in batch
  (`Model.TINY_STREAMING`, `Model.SMALL_STREAMING` (default),
  `Model.MEDIUM_STREAMING`). See `examples/voice/voice-moonshine.py`.
  (PR [#4683](https://github.com/pipecat-ai/pipecat/pull/4683))

- New features for the Vonage WebRTC transport
    - Captions support
    - Individual audio stream subscription support
    - Updated to Vonage Video Connector library v1.0.2
  (PR [#4686](https://github.com/pipecat-ai/pipecat/pull/4686))

- `FunctionSchema` now accepts an optional `handler`. When set, the LLM service
  registers it automatically wherever the schema is advertised in an
  `LLMContext` (or via an `LLMSetToolsFrame`), so no separate
  `register_function` call is needed. This extends the existing
  auto-registration of direct functions to `FunctionSchema`-based tools: the
  advertised tool set stays the single source of truth, so dropping a
  handler-carrying schema unregisters its handler too. A `FunctionSchema`
  without a handler stays advertise-only. Decorate the handler with
  `@tool_options` to override its default call options
  (`cancel_on_interruption`, `timeout_secs`), the same decorator direct
  functions use.
  (PR [#4709](https://github.com/pipecat-ai/pipecat/pull/4709))

- Added `pipecat init`, which makes a project agent-ready by writing a Pipecat
  coding-agent guide (`AGENTS.md` plus a `CLAUDE.md` that imports it) and
  developer guidance (`GETTING_STARTED.md` — MCP setup, how to write a good
  first prompt with a copyable example, what to expect from the session) into
  the project, so an AI coding assistant picks up Pipecat conventions
  automatically and then scaffolds the app with `pipecat create`. Run `pipecat
  init` (prompts for a directory), `pipecat init my-bot`, or `pipecat init .`;
  re-running refreshes `AGENTS.md` while preserving an existing `CLAUDE.md`
  (pass `--force` to overwrite it). The written `AGENTS.md` ends with a
  provenance footer naming the `pipecat-ai` version that wrote it, so a stale
  guide is detectable and refreshable.
  (PR [#4710](https://github.com/pipecat-ai/pipecat/pull/4710))

- Added context carryover support to `AssemblyAISTTService` for Universal-3 Pro
  streaming (`u3-rt-pro`). A new `agent_context` setting seeds the agent's most
  recent reply at connect time, and
  `AssemblyAISTTService.update_agent_context()` updates it mid-session via an
  `UpdateConfiguration` message (no reconnect). Giving the model the agent's
  last reply improves transcription of the user's next turn — short answers,
  spelled-out entities, and similar-sounding words. A
  `previous_context_n_turns` setting controls how many prior entries are
  carried forward (set to `0` to disable carryover entirely). U3 Pro features
  are recognized for the whole `u3-rt-pro` family, including the
  `u3-rt-pro-beta-1` variant.

  - Added `universal-3-5-pro` as a supported `AssemblyAISTTService` model. It
    is recognized as part of the Universal-3 Pro family, so every `u3-rt-pro`
    feature (built-in turn detection, prompting, continuous partials,
    `interruption_delay`, context carryover, and voice focus) applies to it as
    well.

  - Added `voice_focus` and `voice_focus_threshold` settings to
    `AssemblyAISTTService` (Universal-3 Pro models). Set `voice_focus` to
    `"near-field"` or `"far-field"` to isolate the primary voice and suppress
    background noise; `voice_focus_threshold` (0.0–1.0) tunes how aggressively
    background audio is suppressed.
  (PR [#4712](https://github.com/pipecat-ai/pipecat/pull/4712))

- Added the "Add a WebRTC transport for local testing?" option to the Daily
  PSTN and Twilio + Daily SIP scenarios in `pipecat init`, so the generated
  bots can also be run locally with the SmallWebRTC or Daily client.
  (PR [#4715](https://github.com/pipecat-ai/pipecat/pull/4715))

- Realtime and speech-to-speech LLM services that take tools at construction
  now accept a plain list of standard tools (direct functions and/or
  `FunctionSchema` objects), not just a `ToolsSchema` — matching
  `LLMContext(tools=...)`. Applies to `GeminiLiveLLMService` /
  `GeminiLiveVertexLLMService` (`tools=`), `AWSNovaSonicLLMService` (`tools=`),
  `UltravoxRealtimeLLMService` (`one_shot_selected_tools=`), and
  `session_properties.tools` on the `OpenAIRealtimeLLMService` /
  `AzureRealtimeLLMService` / `GrokRealtimeLLMService` /
  `InworldRealtimeLLMService`.
  (PR [#4758](https://github.com/pipecat-ai/pipecat/pull/4758))

- Added `STTService.process_assistant_turn(text)` hook that subclasses can
  override to feed the completed bot reply to a provider-side context carryover
  API. The base implementation is a no-op; `STTService` now handles
  `LLMContextAssistantTurnFrame` and calls this method automatically.
  (PR [#4759](https://github.com/pipecat-ai/pipecat/pull/4759))

- Added `LLMContextAssistantTurnFrame`, broadcast by `LLMAssistantAggregator`
  when a bot turn completes, carrying the aggregated reply text and start
  timestamp.
  (PR [#4759](https://github.com/pipecat-ai/pipecat/pull/4759))

- Added `endpoint_sensitivity` to `SonioxSTTService.Settings`, a float in
  `[-1.0, 1.0]` that controls how aggressively Soniox emits speech endpoints.
  Higher values finalize turns sooner; lower values delay them. Introduced in
  the Soniox v5 model; earlier models reject it.
  (PR [#4772](https://github.com/pipecat-ai/pipecat/pull/4772))

- `DailyTransport` can now publish a `screenAudio` output track, mirroring
  `screenVideo`. Add `"screenAudio"` to `DailyParams.audio_out_destinations`
  (and optionally configure it via `custom_audio_track_params["screenAudio"]`),
  then write audio frames with `transport_destination="screenAudio"`. Requires
  `daily-python>=0.29.0`.
  (PR [#4775](https://github.com/pipecat-ai/pipecat/pull/4775))

- Added an `evals` extra that bundles the `pipecat eval` command (the `cli`
  extra) with the harness's default local, no-API-key models: Kokoro (user-turn
  TTS) and Moonshine (bot-speech transcription). Install `pipecat-ai[evals]` so
  `uv run pipecat eval run` works out of the box. Scaffolded projects (`pipecat
  init`) that enable evals now depend on `pipecat-ai[evals]`.
  (PR [#4776](https://github.com/pipecat-ai/pipecat/pull/4776))

- `RTVIObserver` can now emit raw VAD user speaking events
  (`vad-user-started-speaking` / `vad-user-stopped-speaking`), driven directly
  by the VAD signal and independent of turn finalization (unlike
  `user-started-speaking` / `user-stopped-speaking`, which a turn strategy may
  gate or defer). Enable with
  `RTVIObserverParams(vad_user_speaking_enabled=True)` (off by default), or at
  runtime via `RTVIConfigureObserverFrame`.
  (PR [#4785](https://github.com/pipecat-ai/pipecat/pull/4785))

### Changed

- Migrated all realtime LLM service examples (OpenAI Realtime, Azure Realtime,
  Inworld, Grok/xAI Realtime, Gemini Live, Gemini Live Vertex, AWS Nova Sonic,
  Ultravox) to use `LLMContextAggregatorPair(..., realtime_service_mode=True)`.
  Where examples previously wired `SileroVADAnalyzer` into
  `LLMUserAggregatorParams` as a workaround for missing turn frames, the local
  VAD has been removed; `LLMContextAggregatorPair`'s `realtime_service_mode`
  makes this safe in terms of context-writing. Transcript-logging user-side
  event handlers have moved from `on_user_turn_stopped` to the new
  `on_user_turn_message_added` event, which carries the finalized message text
  (the turn-stopped event fires before the message is finalized in realtime
  service mode). Examples for services without server-side user-turn frames
  (Gemini Live, AWS Nova Sonic, Ultravox) include a comment block explaining
  how to add local VAD if needed. Each base example now also subscribes to
  `on_user_turn_stopped` — active for services that emit server-side user-turn
  frames (OpenAI Realtime, Azure Realtime, Grok, Inworld) and commented-out for
  those that don't (with the same opt-in path as the local-VAD block).
  (PR [#4533](https://github.com/pipecat-ai/pipecat/pull/4533))

- `UserTurnStoppedMessage.content` is now typed `str | None`. In realtime mode
  (`realtime_service_mode=True` on `LLMContextAggregatorPair`) the user message
  isn't finalized at turn-stop time, so `content` is `None`; subscribers
  wanting the finalized text should use the new `on_user_turn_message_added`
  event. Behavior in cascade (STT -> LLM -> STT) pipelines is unchanged.
  (PR [#4533](https://github.com/pipecat-ai/pipecat/pull/4533))

- `SpeechTimeoutUserTurnStopStrategy`, `TurnAnalyzerUserTurnStopStrategy`, and
  `ExternalUserTurnStopStrategy` now accept a `wait_for_transcript: bool =
  True` kwarg. When flipped to `False`, the strategy signals end-of-turn as
  soon as its requirements are met, minus waiting for transcripts — useful when
  you intend to configure local turn detection to drive realtime service
  conversations, where waiting for transcripts is unnecessary latency.
  `LLMContextAggregatorPair` flips this for you when
  `realtime_service_mode=True`.
  (PR [#4533](https://github.com/pipecat-ai/pipecat/pull/4533))

- Updated Smallest AI TTS plugin for Waves v4.0.0 API:
    - New WebSocket endpoint `/waves/v1/tts/live` (previously
      `/waves/v1/{model}/get_speech/stream`)
    - Model is now sent in each message payload instead of the URL, eliminating
      reconnection on model change
    - Updated model names: `lightning_v3.1` and `lightning_v3.1_pro`
      (underscore convention)
    - Added `output_format` setting supporting `pcm`, `mp3`, `wav`, `ulaw`,
      `alaw`
    - Default model changed to `lightning_v3.1_pro` (with `meher` as its
      default voice)
    - **Breaking**: `SmallestTTSModel.LIGHTNING_V2` removed; `consistency`,
      `similarity`, `enhancement` settings removed
  (PR [#4535](https://github.com/pipecat-ai/pipecat/pull/4535))

- ⚠️ RTVI protocol version bumped to `2.0.0`. The `bot-output` message now
  includes `will_be_spoken`, `spoken_status` (`"new"` / `"in-progress"` /
  `"completed"`), `spoken_progress` (accumulated/remaining text), and
  `segment_id` fields. Clients on any `1.x` protocol are still served with the
  legacy format; all other pre-2.x clients are rejected.
  (PR [#4559](https://github.com/pipecat-ai/pipecat/pull/4559))

- `bot_output_transforms` now supports a 4-parameter progress-aware signature:
  `(text, agg_type, accumulated_text, remaining_text) ->
  BotOutputTransformResult`. When called for a progress event,
  `accumulated_text` and `remaining_text` are populated and the transform must
  return a `BotOutputTransformResult` with those fields set, enabling
  word-level transforms on the client side.
  (PR [#4559](https://github.com/pipecat-ai/pipecat/pull/4559))

- Updated `aic-sdk` dependency to `~=2.3.0`. The `AIC_SDK_LICENSE`
  environment variable replaces the previous `AIC_LICENSE_KEY` so the
  variable matches the SDK's canonical name; users must update their
  `.env` files.
  (PR [#4588](https://github.com/pipecat-ai/pipecat/pull/4588))

- Aligned the deprecation docstrings in `LLMUserAggregatorParams` with the
  project's documented convention by removing redundant inline `[DEPRECATED]`
  tags, keeping only the `.. deprecated::` Sphinx directive.
  (PR [#4592](https://github.com/pipecat-ai/pipecat/pull/4592))

- `AzureSTTService` now marks final transcripts as finalized. Azure's
  `RecognizedSpeech` event is by definition the final recognition for an
  utterance, so the emitted `TranscriptionFrame` carries `finalized=True`. This
  lets downstream user-turn stop strategies (e.g. `SpeechTimeoutUserTurnStop`)
  take their finalized fast-path instead of waiting for VAD events that may
  never arrive on short replies.
  (PR [#4620](https://github.com/pipecat-ai/pipecat/pull/4620))

- ⚠️ The `mem0` extra now requires `mem0ai>=2,<3`. `Mem0MemoryService` was
  updated for the mem0 2.0.0 breaking changes: entity IDs
  (`user_id`/`agent_id`/`run_id`) are now passed via `filters=` to the local
  client (top-level kwargs raise `ValueError` in mem0 2.x), and the removed
  `version`/`output_format` parameters are no longer sent to the cloud client.
  Note that mem0 2.0.0 also flips the `rerank` default from `True` to `False`
  and makes `add()` async server-side (stored memories are queryable once
  processed).
  (PR [#4626](https://github.com/pipecat-ai/pipecat/pull/4626))

- `GradiumSTTService` now defaults `delay_in_frames` to `12` (960ms) instead of
  leaving it unset (which used the server default of 10/800ms). The higher
  default allows more context for improved transcription accuracy. Set
  `delay_in_frames` explicitly to `7`-`8` for faster responses.
  (PR [#4632](https://github.com/pipecat-ai/pipecat/pull/4632))

- `GradiumSTTService` has an updated `ttfs_p99_latency` value of 0.62 seconds.
  (PR [#4632](https://github.com/pipecat-ai/pipecat/pull/4632))

- Bumped `pipecat-ai-prebuilt` to 1.0.2 in the `runner` extra, updating the
  prebuilt client UI served by the development runner.
  (PR [#4634](https://github.com/pipecat-ai/pipecat/pull/4634))

- ⚠️ Changed the default of `TTSSpeakFrame.append_to_context` from `None` to
  `True`. The old `None` behavior was situation-dependent and hard to reason
  about: the spoken text always reached the assistant aggregator's buffer, but
  whether it was committed to the LLM context depended on what surrounded the
  frame — committed when the frame was inside an assistant response or
  immediately followed by one, but silently discarded when it was standalone and
  followed by a user turn (the interruption cleared the buffer before anything
  flushed it). `True` is a predictable default: programmatically-spoken text is
  recorded in the context unless you opt out with `append_to_context=False`.
  `BusTTSSpeakMessage.append_to_context` now defaults to `True` to match.
  (PR [#4642](https://github.com/pipecat-ai/pipecat/pull/4642))

- Switched the `aws` extra from `aioboto3` to `aiobotocore`. Pipecat only uses
  the low-level client API, and `aiobotocore` is the async library that
  `aioboto3` wraps, so depending on it directly drops an unnecessary wrapper
  layer. AWS service initialization now uses
  `aiobotocore.session.get_session()` and `session.create_client(...)`; public
  APIs and credential resolution are unchanged.
  (PR [#4643](https://github.com/pipecat-ai/pipecat/pull/4643))

- `websockets` is now a core dependency of `pipecat-ai` instead of the
  `websockets-base` optional extra. The `websockets-base` extra has been
  removed; service extras that used to pull it in (Cartesia, Deepgram,
  ElevenLabs, OpenAI, Google, and others) still work unchanged, and
  `websockets` is now always installed. If you previously installed
  `pipecat-ai[websockets-base]` directly, just drop the extra since `pip
  install pipecat-ai` now includes it.
  (PR [#4658](https://github.com/pipecat-ai/pipecat/pull/4658))

- Renamed the `@tool` decorator's `timeout` argument to `timeout_secs`,
  matching `register_function()`. `timeout` still works as a deprecated alias
  and will be removed in a future version.
  (PR [#4671](https://github.com/pipecat-ai/pipecat/pull/4671))

- `WhisperSTTService`'s `Model` and `MLXModel` are now `StrEnum`, so a member
  is the string itself (e.g. `Model.TINY == "tiny"`). Passing a
  `Model`/`MLXModel` member or a plain string both keep working.
  (PR [#4684](https://github.com/pipecat-ai/pipecat/pull/4684))

- Bumped the `daily` extra's `daily-python` dependency to `>=0.29.1,<1`.
  (PR [#4685](https://github.com/pipecat-ai/pipecat/pull/4685))

- `LLMWorker` now enables the worker's automatic RTVI support when it is not
  bridged (`bridged=None`), so a standalone `LLMWorker` driving its own
  transport gets the `RTVIProcessor`/`RTVIObserver` pair like any
  `PipelineWorker`. Bridged child workers keep RTVI disabled, since the
  transport worker owns the client-facing RTVI machinery.
  (PR [#4690](https://github.com/pipecat-ai/pipecat/pull/4690))

- Removed the `asyncio.sleep(0)` workarounds that let a just-created timer task
  start before a possible immediate cancellation. `TaskManager.create_task()`
  now cleans up never-started coroutines centrally, so the yields served no
  purpose.
  (PR [#4692](https://github.com/pipecat-ai/pipecat/pull/4692))

- Worker frames (e.g. `EndWorkerFrame`) should now be pushed downstream with a
  plain `push_frame(frame)`, so frames queued ahead of them are flushed before
  the worker acts on them. Pushing them upstream still works.
  (PR [#4705](https://github.com/pipecat-ai/pipecat/pull/4705))

- `register_function` now reads a handler's call options
  (`cancel_on_interruption`, `timeout_secs`) from its `@tool_options` decorator
  when they aren't passed explicitly, matching how direct functions resolve
  them (explicit argument > `@tool_options` > default). Previously the
  decorator was ignored on this path.
  (PR [#4709](https://github.com/pipecat-ai/pipecat/pull/4709))

- `BaseLLMAdapter.from_standard_tools` now raises `UserWarning` instead of
  `DeprecationWarning` when built-in tools can't be injected because the
  supplied tools aren't a `ToolsSchema` — it advises about the tools format and
  is not a deprecation.
  (PR [#4726](https://github.com/pipecat-ai/pipecat/pull/4726))

- Deprecated classes and functions are now marked with the PEP 702
  `@deprecated` decorator, so type checkers and IDEs (pyright/Pylance
  `reportDeprecated`, mypy's `deprecated` error code) flag and strike through
  deprecated usages statically. Several deprecated classes that previously
  emitted no runtime warning now raise `DeprecationWarning` when used, and
  deprecation messages now state a concrete removal version (e.g. `2.0.0`)
  instead of "a future release".
  (PR [#4726](https://github.com/pipecat-ai/pipecat/pull/4726))

- `pipecat create` now infers `--bot-type` from the chosen transports in
  non-interactive mode, so the flag is optional: a bot is `telephony` when any
  transport is a telephony transport (twilio, telnyx, plivo, exotel,
  daily_pstn, twilio_daily_sip) and `web` otherwise. Pass `--bot-type`
  explicitly to override (it's still validated and cross-checked against the
  transports); the interactive wizard is unchanged.
  (PR [#4735](https://github.com/pipecat-ai/pipecat/pull/4735))

- Realtime LLM services now auto-register the handlers bundled on the tools
  passed at construction time, so a separate `register_function()` call is no
  longer needed — matching how context-advertised tools (a direct function, or
  a `FunctionSchema` with its `handler` set) already register. Applies to
  `GeminiLiveLLMService` / `GeminiLiveVertexLLMService` (`tools=`),
  `UltravoxRealtimeLLMService` (`one_shot_selected_tools=`),
  `AWSNovaSonicLLMService` (`tools=`), and `OpenAIRealtimeLLMService` /
  `AzureRealtimeLLMService` / `GrokRealtimeLLMService` /
  `InworldRealtimeLLMService` (`session_properties.tools`).
  (PR [#4758](https://github.com/pipecat-ai/pipecat/pull/4758))

- Updated `SonioxSTTService` default model from `stt-rt-v4` to `stt-rt-v5`.
  (PR [#4772](https://github.com/pipecat-ai/pipecat/pull/4772))

- The Kokoro TTS model cache moved to `~/.cache/pipecat/kokoro-onnx`
  (previously `~/.cache/kokoro-onnx`), so Pipecat's cached files live under a
  single namespaced directory.
  (PR [#4776](https://github.com/pipecat-ai/pipecat/pull/4776))

### Deprecated

- Deprecated the 2-parameter `bot_output_transforms` signature `(text,
  agg_type) -> str`. Transforms using it will still work but emit a
  `DeprecationWarning` at registration time. Update to the 4-parameter
  signature `(text, agg_type, accumulated_text, remaining_text) ->
  BotOutputTransformResult` to support word-level progress transforms.
  (PR [#4559](https://github.com/pipecat-ai/pipecat/pull/4559))

- ⚠️ Deprecated `AICVADAnalyzer` (`pipecat.audio.vad.aic_vad`) and
  `AICFilter.create_vad_analyzer()`. Both are tied to `AICFilter`'s
  model-internal VAD path. Use `AICQuailVADAnalyzer` instead — the standalone
  Quail VAD 2.0 model is the noise-robust VAD differentiator going forward.
  Both surfaces will be removed in Pipecat 1.6.0 (breaking change shipped in a
  minor release, per maintainer guidance for plugins).
  (PR [#4588](https://github.com/pipecat-ai/pipecat/pull/4588))

- The single-argument `connection_callback(connection)` signature for
  `WhatsAppClient.handle_webhook_request` is deprecated. Update callbacks to
  accept `(connection, call: WhatsAppConnectCall)` to receive call metadata
  alongside the WebRTC connection. The old signature still works but emits a
  `DeprecationWarning`.
  (PR [#4622](https://github.com/pipecat-ai/pipecat/pull/4622))

- Deprecated `Mem0MemoryService.InputParams.api_version`. It is no longer used
  — mem0 2.0.0 removed the `api_version`/`output_format` parameters from the
  client. Setting it now emits a `DeprecationWarning`.
  (PR [#4626](https://github.com/pipecat-ai/pipecat/pull/4626))

- Deprecated passing `append_to_context=None` to `TTSSpeakFrame` (and
  `BusTTSSpeakMessage`). `None` is no longer a supported value: it is coerced to
  `True` with a warning and will be unsupported in a future release. Pass `True`
  or `False` explicitly. See the corresponding "Changed" entry for the full
  rationale behind the new `True` default.
  (PR [#4642](https://github.com/pipecat-ai/pipecat/pull/4642))

- Deprecated `LLMService.register_direct_function()` /
  `unregister_direct_function()` and `LLMSwitcher.register_direct_function()`.
  Advertise direct functions in `LLMContext(tools=[...])` or via an
  `LLMSetToolsFrame` instead — handlers are registered and unregistered
  automatically. These will be removed in a future version.
  (PR [#4671](https://github.com/pipecat-ai/pipecat/pull/4671))

- ⚠️ Deprecated `TaskFrame`, `TaskSystemFrame`, `EndTaskFrame`,
  `StopTaskFrame`, `CancelTaskFrame` and `InterruptionTaskFrame`. Use
  `WorkerFrame`, `WorkerSystemFrame`, `EndWorkerFrame`, `StopWorkerFrame`,
  `CancelWorkerFrame` and `InterruptionWorkerFrame` instead, matching the
  `PipelineWorker` naming. The old names remain as `isinstance`-compatible
  aliases that emit a `DeprecationWarning` on construction.
  (PR [#4705](https://github.com/pipecat-ai/pipecat/pull/4705))

- Renamed `WebsocketServerTransport` to `SingleClientWebsocketServerTransport`
  to make it explicit that the server handles a single client at a time. The
  supporting `WebsocketServerParams`, `WebsocketServerCallbacks`,
  `WebsocketServerInputTransport`, and `WebsocketServerOutputTransport` classes
  were renamed with the same `SingleClient` prefix. The old names remain as
  deprecated aliases and will be removed in 2.0.0.
  (PR [#4774](https://github.com/pipecat-ai/pipecat/pull/4774))

### Fixed

- Fixed output image resizing for generated images when video output dimensions
  differ from the source image size by consistently using Pillow pixel modes
  instead of encoded formats.
  (PR [#4483](https://github.com/pipecat-ai/pipecat/pull/4483))

- Fixed a benign `ERROR` log line emitted by `UltravoxRealtimeLLMService`
  during client-driven teardown. Adds an exception catch which guards the
  disconnecting case.
  (PR [#4519](https://github.com/pipecat-ai/pipecat/pull/4519))

- Fixed `InworldRealtimeLLMService` not supporting manual-mode turn detection
  (`session_properties.audio.input.turn_detection=None`). Previously
  `_handle_user_stopped_speaking` and `_handle_interruption` assumed Inworld's
  server-side VAD handled commit/cancel/response.create automatically and were
  no-ops on the client side. In manual mode the server doesn't, so
  local-VAD-driven turns stalled: the bot never responded after the user
  stopped speaking, and interruptions didn't cancel the in-flight response.
  Wire the explicit `InputAudioBufferCommitEvent` + `ResponseCreateEvent` on
  user-stopped-speaking and `InputAudioBufferClearEvent` +
  `ResponseCancelEvent` on interruption, gated on a new
  `_is_manual_turn_detection()` check (mirroring the pattern in
  `OpenAIRealtimeLLMService`).
  (PR [#4533](https://github.com/pipecat-ai/pipecat/pull/4533))

- `InworldRealtimeLLMService` and `GrokRealtimeLLMService` no longer broadcast
  `UserStartedSpeakingFrame`/`UserStoppedSpeakingFrame` when configured for
  manual (locally-driven) turn detection. Both services' server-side
  speech-started/stopped events fire in manual mode too, but in that setup turn
  frames are expected to come from local turn detection (e.g. a `vad_analyzer`
  in `LLMUserAggregatorParams`) — without the gate, the services were
  broadcasting alongside the locally-emitted frames, producing duplicate
  `on_user_turn_*` events. OpenAI Realtime was already correct here: its server
  doesn't fire speech events in manual mode at all.
  (PR [#4533](https://github.com/pipecat-ai/pipecat/pull/4533))

- Fixed Ultravox Realtime not surfacing server-side interruption. The server
  sends a `playback_clear_buffer` message when the user interrupts the bot
  mid-speech, instructing clients to drop buffered output audio; this was
  previously unhandled, so `BaseOutputTransport` kept playing the buffered
  audio and the bot kept talking past the interruption. Ultravox now broadcasts
  `InterruptionFrame` on `playback_clear_buffer`. This was previously masked by
  enabling local VAD on the user aggregator, which generated
  `UserStartedSpeakingFrame` and triggered the aggregator-side interruption
  path; the fix makes the behavior correct without local VAD as a workaround.
  (PR [#4533](https://github.com/pipecat-ai/pipecat/pull/4533))

- Fixed `GrokRealtimeLLMService` stalling the conversation when Grok returns an
  error in response to a `response.cancel` event sent while no response is
  active on the server. This happens routinely in manual-turn-detection mode:
  when the user starts speaking after the bot has finished, Pipecat broadcasts
  an `InterruptionFrame` and the service sends `ResponseCancelEvent`, which
  Grok rejects with `"Cancellation failed: no active response found"`. The
  existing error-suppression list only matched OpenAI's
  `response_cancel_not_active` / `conversation_already_has_active_response`
  error codes, but Grok uses different codes for the same conditions — so the
  error fell through to the fatal-error path and exited the WebSocket receive
  loop, preventing any further server events from being processed. The
  suppression now also matches on the error message substring (`"no active
  response"`, `"already has an active response"`), so these benign races get
  logged at debug and the receive loop keeps running.
  (PR [#4533](https://github.com/pipecat-ai/pipecat/pull/4533))

- Fixed AWS Nova Sonic not surfacing server-side interruption. When the user
  interrupted the bot mid-response, the `INTERRUPTED` stop reason was
  acknowledged internally but no `InterruptionFrame` was emitted, so
  `BaseOutputTransport` kept draining its audio buffer and the bot kept talking
  past the interruption. Nova Sonic now broadcasts `InterruptionFrame` on both
  `INTERRUPTED` paths (text-stage and audio-stage). This was previously masked
  by enabling local VAD on the user aggregator, which generated
  `UserStartedSpeakingFrame` and triggered the aggregator-side interruption
  path; the fix makes the behavior correct without local VAD as a workaround.
  (PR [#4533](https://github.com/pipecat-ai/pipecat/pull/4533))

- Fixed pipeline shutdown hanging on LiveKit when the remote peer disconnected
  mid-stream. The trailing `audio_out_end_silence_secs` write is now bounded by
  a timeout.
  (PR [#4578](https://github.com/pipecat-ai/pipecat/pull/4578))

- Fixed the start of user speech being clipped from transcripts when
  `GeminiLiveLLMService` is configured for locally-driven turns (server-side
  VAD disabled). The problem was that any audio sent up to Gemini Live before
  sending `activity_start` (sent when user turn start is confirmed) seemed to
  get discarded; the service now replays (sends to Gemini Live) a short audio
  "pre-roll" right after `activity_start`, so the onset is preserved.
  (PR [#4597](https://github.com/pipecat-ai/pipecat/pull/4597))

- Fixed the start of user speech being clipped from transcripts when
  `OpenAIRealtimeLLMService` is configured for locally-driven turns
  (server-side turn detection disabled). The problem was that the speech onset
  already sent to OpenAI got discarded when the service cleared its input audio
  buffer on barge-in (which it does when the user turn start is confirmed); the
  service now replays (re-appends to the input audio buffer) a short audio
  "pre-roll" right after the clear, so the onset is preserved.
  (PR [#4599](https://github.com/pipecat-ai/pipecat/pull/4599))

- 422 validation errors now log the full error details and raw request body for
  all transports (WhatsApp, WebRTC, telephony, etc.), making malformed payloads
  easier to debug. Previously this logging only applied to WhatsApp routes.
  (PR [#4622](https://github.com/pipecat-ai/pipecat/pull/4622))

- Fixed `InworldTTSService` logging a spurious "no websocket connected, will
  try to reconnect" warning and firing a redundant second reconnect when the
  initial connection attempt failed. The service now returns an `ErrorFrame`
  immediately if the websocket is unavailable after `_connect()`, matching the
  behaviour of `ElevenLabsTTSService`.
  (PR [#4635](https://github.com/pipecat-ai/pipecat/pull/4635))

- Fixed `SarvamTTSService` (WebSocket) emitting `BotStoppedSpeakingFrame` late.
  The service never produced a `TTSStoppedFrame` on synthesis completion, so
  end-of-turn was detected only by the `stop_frame_timeout_s` idle timer,
  causing `BotStoppedSpeakingFrame` to lag the actual end of audio by up to
  that timeout (especially for short utterances or a raised
  `stop_frame_timeout_s`). The service now requests Sarvam's completion event
  (`send_completion_event`) and emits `TTSStoppedFrame` as soon as the `final`
  event arrives, so the bot-stopped-speaking event tracks the end of audio. The
  idle timeout remains as a fallback.
  (PR [#4639](https://github.com/pipecat-ai/pipecat/pull/4639))

- Fixed a spurious `RuntimeWarning: coroutine '...' was never awaited` emitted
  by `TaskManager.create_task()` when a task is cancelled before its coroutine
  starts running. The wrapper now closes the un-started coroutine on
  cancellation, so the warning no longer fires. This surfaced, for example,
  when combining `TurnAnalyzerUserTurnStopStrategy` with another stop strategy
  that force-completes the turn (cancelling the analyzer's timeout task before
  it ran), and when a function call is cancelled by a user-turn-start
  interruption race (the `LLMService._run_function_call` warning,
  [#4339](https://github.com/pipecat-ai/pipecat/issues/4339)). A local `await
  asyncio.sleep(0)` workaround in `_run_function_call` that existed only to
  dodge this warning has been removed now that it is handled centrally. The
  turn/cancellation behavior was already correct; only the noisy warning is
  removed.
  (PR [#4644](https://github.com/pipecat-ai/pipecat/pull/4644))

- Fixed `LiveKitTransport` leaking audio/video stream readers when a track is
  unsubscribed: the owned `rtc.AudioStream`/`rtc.VideoStream` and its producer
  task are now closed and cancelled on unsubscribe (and on a re-subscribe for
  the same participant), so a client republishing its mic (e.g. mute/unmute or
  text↔voice toggles) no longer accumulates concurrent producers that
  interleave audio into the shared queue and silence downstream STT.
  (PR [#4650](https://github.com/pipecat-ai/pipecat/pull/4650))

- Fixed `TTSService` emitting a second `LLMFullResponseEndFrame` (with a new
  id) at the end of an audio context when `push_text_frames` is `False`, which
  caused `RTVIObserver` to send a duplicate `bot-llm-stopped` message per LLM
  response. The original end frame received in `process_frame` is now held per
  `context_id` and re-pushed, preserving its id.
  (PR [#4653](https://github.com/pipecat-ai/pipecat/pull/4653))

- Fixed a frame-ordering race in bridged workers: frames received from the
  `WorkerBus` were pushed directly into the pipeline from the bus edge, so they
  could interleave with (or overtake) frames the worker had queued itself via
  `queue_frame()`/`queue_frames()`. A bus inbound frame (e.g. an
  `LLMContextFrame` from a concurrent user input) could reach the LLM in the
  middle of a multi-frame update such as a flow's `set_node`
  (`LLMMessagesUpdateFrame` + `LLMSetToolsFrame`), generating against the
  previous node's context. Bus inbound frames are now serialized through the
  worker's frame queue, so both paths share one FIFO.
  (PR [#4656](https://github.com/pipecat-ai/pipecat/pull/4656))

- Fixed interruption handling for standalone
  `TTSSpeakFrame(append_to_context=True)` utterances (those not part of an LLM
  response). Previously, when the user interrupted such an utterance:
    - `on_assistant_turn_stopped` didn't fire
    - partially-spoken text wasn't recorded to the context (for TTS services
      that support word timestamps)

  The problem was that these utterances have no `LLMFullResponseStartFrame`
  to open the assistant turn, so there was no open turn for the interruption to
  stop. The assistant aggregator now uses a new
  `TTSStartedFrame.append_to_context` to open the turn when the utterance
  begins.

  As a result of this fix, `on_assistant_turn_started` timing is improved for
  standalone `TTSSpeakFrame` utterances: the event now fires at the start
  rather than at the end.
  (PR [#4665](https://github.com/pipecat-ai/pipecat/pull/4665))

- Fixed `OpenAIResponsesHttpLLMService` raising `'NoneType' object has no
  attribute 'cached_tokens'` on every turn when used with a custom `base_url`
  pointing at a third-party Responses API server that omits the OpenAI-specific
  `input_tokens_details` / `output_tokens_details` sub-objects. Token usage
  parsing now tolerates any field the server omits — the SDK's lenient
  streaming decoder leaves these as `None` whether it's a top-level count
  (`input_tokens` / `output_tokens` / `total_tokens`), a missing detail
  sub-object, or a missing field inside one — and falls back to `0` in each
  case, matching the WebSocket `OpenAIResponsesLLMService` variant.
  (PR [#4667](https://github.com/pipecat-ai/pipecat/pull/4667))

- Fixed importing `pipecat.services.whisper.stt` failing on non-macOS platforms
  when the `mlx-whisper` extra happened to be installed: `mlx_whisper` is now
  only imported on macOS (it's Apple-Silicon only, and elsewhere the package
  can be installed but unloadable, e.g. a missing `libmlx.so`).
  `WhisperSTTServiceMLX` still imports it lazily when actually used.
  (PR [#4684](https://github.com/pipecat-ai/pipecat/pull/4684))

- Fixed `SambaNovaLLMService` failing every completion with its default model:
  SambaNova Cloud removed `Llama-4-Maverick-17B-128E-Instruct`, so the default
  is now `Meta-Llama-3.3-70B-Instruct`.
  (PR [#4687](https://github.com/pipecat-ai/pipecat/pull/4687))

- Fixed `NebiusLLMService` function calls never executing with its default
  model: Nebius streams `openai/gpt-oss-120b` tool calls with a broken final
  fragment (`index=1` on a single call), so the default is now
  `Qwen/Qwen3-30B-A3B-Instruct-2507`, which streams correctly.
  (PR [#4688](https://github.com/pipecat-ai/pipecat/pull/4688))

- Fixed a worker-handoff race where `activate_worker(deactivate_self=True)`
  left both workers briefly active: the caller's `active` flag only flipped
  when its own deactivate message came back over the bus, so the handoff target
  could activate first and both workers re-broadcast each other's frames
  (duplicate tool round-trips in the LLM context). The caller now deactivates
  synchronously before publishing the activate message.
  (PR [#4691](https://github.com/pipecat-ai/pipecat/pull/4691))

- Fixed `AzureTTSService` producing no audio when running in a pipeline without
  an output transport (e.g. headless/offline setups). Audio chunks arrive from
  the Speech SDK on native threads, and the cross-thread queue hand-off didn't
  wake an otherwise-idle event loop; the service now marshals those puts onto
  the loop, so audio is delivered regardless of loop activity.
  (PR [#4703](https://github.com/pipecat-ai/pipecat/pull/4703))

- Fixed a regression from #4654 where unregistering a tool's handler on its own
  — via `unregister_function` / `unregister_direct_function`, without changing
  the advertised tool set — was silently undone, because the next
  `LLMContextFrame` re-registered the handler from the still-advertised tool (a
  "zombie"). An explicitly unregistered handler now stays unregistered while
  its tool remains advertised (so calls hit the missing-handler recovery and
  the model learns to stop), and is restored only by registering it again, or
  by re-advertising the tool (removing it from the advertised set, then adding
  it back).
  (PR [#4709](https://github.com/pipecat-ai/pipecat/pull/4709))

- Fixed `LLMSwitcher.register_direct_function` overriding a direct function's
  `@tool_options` call options. Its `cancel_on_interruption` defaulted to
  `True` and was forwarded to each member LLM as an explicit value, so a
  `@tool_options(cancel_on_interruption=False)` handler was ignored. It now
  defaults to `None` and follows the same explicit-arg > `@tool_options` >
  default fallback as `LLMService.register_direct_function` (the fallback was
  added there in #4654 but not mirrored on the switcher).
  (PR [#4709](https://github.com/pipecat-ai/pipecat/pull/4709))

- Fixed an audible 200-300 ms gap in audio mixer output (e.g. `SoundfileMixer`
  background sound) on every interruption. The output transport now keeps the
  audio task running and drains the queue instead of cancelling and recreating
  it when a mixer is active.
  (PR [#4714](https://github.com/pipecat-ai/pipecat/pull/4714))

- Fixed `pipecat init` silently ignoring the "Enable evals?" option for the
  Daily PSTN Dial-out and Twilio + Daily SIP scenarios. Generated bots for
  these scenarios can now be driven with `pipecat eval` (`-t eval`): they fall
  back to `create_transport()` when the request body carries no `room_url`,
  while the production flow (room and call settings arriving from `server.py`)
  is unchanged. In local runs the dial-out and Twilio call-forwarding machinery
  is skipped and the bot stays silent until spoken to.
  (PR [#4715](https://github.com/pipecat-ai/pipecat/pull/4715))

- Fixed `FastAPIWebsocketTransport` stalling pipeline shutdown for ~10s when
  the client's WebSocket is half-closed (e.g. a telephony call already torn
  down on the provider's side, leaving the media-streams socket open at the TCP
  layer but unresponsive). `FastAPIWebsocketClient.disconnect()` awaited
  `websocket.close()` unbounded, so it blocked on the ASGI server's
  close-handshake timeout, delaying `EndFrame` propagation and the whole
  pipeline teardown. The close handshake is now bounded by a new
  `FastAPIWebsocketParams.ws_close_timeout` (default 0.5s): the close is still
  initiated, but `disconnect()` waits at most that long for the peer's
  acknowledgment before letting shutdown proceed. Increase `ws_close_timeout`
  for high-latency peers that need longer to complete a graceful close.
  (PR [#4723](https://github.com/pipecat-ai/pipecat/pull/4723))

- Fixed `NvidiaLLMService` reasoning streams so interrupted or early-cancelled
  responses clean up correctly and do not leak buffered thought content or
  leave the wrapped stream open.
  (PR [#4743](https://github.com/pipecat-ai/pipecat/pull/4743))

- Fixed an issue in `AggregatedFrameSequencer` where delayed word-timestamps
  from an interrupted (cleared) TTS context could be emitted as passthrough
  `TTSTextFrame`s with `append_to_context=True`, interleaving stale words into
  the next turn's transcript (observed with Inworld TTS in `ASYNC` mode). Words
  for an unknown or cleared context are now dropped instead of corrupting the
  active turn.
  (PR [#4751](https://github.com/pipecat-ai/pipecat/pull/4751))

- Fixed `RimeTTSService.SPELL()` and `RimeTTSService.PAUSE_TAG()` helpers,
  which are now static methods. Previously they were defined as instance
  methods without a `self` parameter, so calling them on a service instance
  bound the instance to the first argument and produced incorrect output.
  (PR [#4755](https://github.com/pipecat-ai/pipecat/pull/4755))

- Fixed `SingleClientWebsocketServerTransport` (formerly
  `WebsocketServerTransport`) so that a new client connection no longer
  disconnects the client that is already connected. While a client is
  connected, new connection attempts are now rejected with a warning. The
  active connection's reference is cleared when the client disconnects or the
  connection fails, so a new client can connect afterwards.
  (PR [#4774](https://github.com/pipecat-ai/pipecat/pull/4774))

- Fixed `DeepgramSageMakerSTTService` raising `TypeError` on construction. Its
  default settings still passed `vad_events`, which was removed from
  `DeepgramSTTService.Settings`, so the service (and the
  `voice-deepgram-sagemaker` example) crashed on instantiation.
  (PR [#4786](https://github.com/pipecat-ai/pipecat/pull/4786))

### Security

- Added optional HMAC token authentication for WebSocket connections in the
  development runner. Set `PIPECAT_WEBSOCKET_AUTH=token` (or pass `--ws-auth
  token`) to require clients to call `POST /start` and obtain a short-lived
  signed session token before connecting. Tokens are one-time use and expire
  after 5 minutes.
    - Clients can supply the token via `Authorization: Bearer <token>` header,
      `?token=<token>` query parameter, or URL path segment (`/ws/<token>`,
      `/ws-client/<token>`) — the path form is recommended for telephony providers
      like Twilio.
    - Both the telephony WebSocket (`/ws`) and plain WebSocket (`/ws-client`)
      endpoints are protected. Connections with invalid, expired, or replayed
      tokens are rejected with WebSocket close code 4003.
  (PR [#4660](https://github.com/pipecat-ai/pipecat/pull/4660))

- Added origin restriction support to `WebsocketServerTransport`,
  `FastAPIWebsocketTransport`, and the development runner to mitigate
  Cross-Site WebSocket Hijacking (CSWSH). When `allowed_origins` is configured,
  connections with a missing or disallowed `Origin` header are rejected before
  the WebSocket handshake completes.
    - `WebsocketServerParams` and `FastAPIWebsocketParams` gain an
      `allowed_origins: list[str]` field. `FastAPIWebsocketTransport` raises
      `ValueError` at construction time if the origin is not allowed.
    - The runner gains `--allowed-origins` CLI flag and
      `PIPECAT_ALLOWED_ORIGINS` environment variable (comma-separated). Both
      also control the transport params default, so a single env var covers all
      WebSocket transports uniformly.
    - Default is empty (allow all) — no behaviour change for existing
      deployments.
  (PR [#4704](https://github.com/pipecat-ai/pipecat/pull/4704))

### Other

- Pinned `vite` to `^8.0.16` (from `^8`) in the `pipecat create` client
  templates and the UI-worker examples.
  (PR [#4767](https://github.com/pipecat-ai/pipecat/pull/4767))

## [1.3.0] - 2026-05-28

### Added

- Pipecat pipelines are multi-agent compatible by default. The new multi-agent
  framework (`pipecat.workers`) turns every `PipelineWorker` (previously
  `PipelineTask`) into a peer on a shared bus that passes typed messages,
  dispatches `@job` work, and coordinates with siblings, while existing
  single-pipeline code keeps running untouched. `examples/multi-worker/` ships
  ready-to-run patterns: LLM handoff, parallel debate, sidecar code assistants
  and hardware controllers, distributed deployments over Redis or PGMQ,
  point-to-point WebSocket proxies, and UI workers driving a web client over
  RTVI.  (PR [#4493](https://github.com/pipecat-ai/pipecat/pull/4493))

- Added `UIWorker` (`pipecat.workers.ui`): an LLM worker that observes and
  drives a client web UI over the RTVI UI channel — for voice agents that act
  on what the user is looking at. It reads the page's accessibility snapshots,
  routes client UI events to `@ui_event` handlers, drives the page with UI
  commands (`scroll_to`, `highlight`, `select_text`, `click`,
  `set_input_value`), and answers screen-grounded questions. `PipelineWorker`
  connects it to the client automatically when RTVI is enabled — no extra
  wiring.
    - A voice agent delegates a turn via the built-in `respond` job; the worker
      returns an answer for the voice LLM to speak, or speaks it verbatim through
      the agent's TTS with `respond_to_job(answer, tts_speak=True)`.
    - `ReplyToolMixin` provides a ready-made `reply` tool (a spoken answer plus
      the standard UI actions).
    - `ui_job_group(...)` fans work out to peer workers, surfaced to the client
      as cancellable progress cards.
    - `UI_STATE_PROMPT_GUIDE` is drop-in system-prompt text that teaches the
       LLM the `<ui_state>` wire format.
  (PR [#4540](https://github.com/pipecat-ai/pipecat/pull/4540))

- Added `VonageVideoConnectorTransport`, a new transport integration for
  real-time Vonage WebRTC sessions using the Vonage Video Connector library.
  (PR [#4052](https://github.com/pipecat-ai/pipecat/pull/4052))

- Added `InceptionLLMService` for Inception's Mercury 2 diffusion reasoning
  model, with support for `reasoning_effort` and `realtime` settings.
  (PR [#4423](https://github.com/pipecat-ai/pipecat/pull/4423))

- Added plain WebSocket transport support to the development runner. Bots can
  now accept connections from non-telephony WebSocket clients (e.g., browser
  apps using protobuf framing) via the `/ws-client` endpoint alongside other
  transports.
  (PR [#4442](https://github.com/pipecat-ai/pipecat/pull/4442))

- Added `GET /status` endpoint to the development runner that reports which
  transports the running instance accepts (all by default, or the single
  transport passed via `-t`).
  (PR [#4442](https://github.com/pipecat-ai/pipecat/pull/4442))

- Added support for the Rime `coda` TTS model to `RimeTTSService` and
  `RimeHttpTTSService`. The `temperature`, `top_p`, and `repetition_penalty`
  settings are not used by `coda`. Also added a `timeScaleFactor` setting (for
  the `arcana` and `coda` models) to both services — values above 1.0 slow down
  audio playback; values below 1.0 speed it up.
  (PR [#4511](https://github.com/pipecat-ai/pipecat/pull/4511))

- Added `max_endpoint_delay_ms` to `SonioxSTTService.Settings`, controlling the
  maximum delay (500-3000 ms) before endpoint detection finalizes a turn.
  (PR [#4521](https://github.com/pipecat-ai/pipecat/pull/4521))

- Added `LLMService.append_system_instruction(...)`: append durable text to a
  service's system instruction so it's included on every inference and survives
  context resets.
  (PR [#4540](https://github.com/pipecat-ai/pipecat/pull/4540))

- Added `CartesiaTurnsSTTService` for streaming speech-to-text against the
  Cartesia Streaming ASR v2 (Ink-2) turn-based WebSocket endpoint
  (`/stt/turns/websocket`). The server drives turn boundaries via `turn.start`
  / `turn.update` / `turn.end` messages, which the service translates into
  `UserStartedSpeakingFrame`, finalized `TranscriptionFrame`, and
  `UserStoppedSpeakingFrame`. Eager end-of-turn predictions and turn resumes
  (`turn.eager_end` and `turn.resume`) are surfaced via the `on_turn_eager_end`
  and `on_turn_resume` event handlers.
  (PR [#4552](https://github.com/pipecat-ai/pipecat/pull/4552))

- Added the `STTService.supports_ttfs` property, which subclasses can override
  to return `False` when TTFS doesn't apply to their architecture (e.g.
  turn-based STTs where the server defines turn boundaries). When `False`,
  `STTMetadataFrame` is broadcast with `ttfs_p99_latency=0.0` and the
  "ttfs_p99_latency not set" warning is suppressed.
  (PR [#4585](https://github.com/pipecat-ai/pipecat/pull/4585))

### Changed

- ⚠️ The development runner now supports all transports (WebRTC, Daily,
  telephony, plain WebSocket) simultaneously from a single server. The `/start`
  endpoint accepts a `"transport"` field to select the transport per-request;
  omitting `-t` at startup enables all transports instead of defaulting to
  WebRTC. The Daily browser-redirect route moved from `GET /` to `GET /daily`.
  (PR [#4442](https://github.com/pipecat-ai/pipecat/pull/4442))

- Changed the default model for `RimeTTSService` and `RimeHttpTTSService` from
  `arcana` to `coda`. Code that relied on the implicit default should set
  `model="arcana"` explicitly to preserve previous behavior.
  (PR [#4511](https://github.com/pipecat-ai/pipecat/pull/4511))

- OpenRouter LLM service now defaults to `openai/gpt-4.1`.
  (PR [#4513](https://github.com/pipecat-ai/pipecat/pull/4513))

- OpenRouter LLM requests now convert `developer` messages to `user` messages
  by default for broader model compatibility. Override this by subclassing
  `OpenRouterLLMService` or setting `llm.supports_developer_role = True` for
  models that support the `developer` role.
  (PR [#4513](https://github.com/pipecat-ai/pipecat/pull/4513))

- `SonioxSTTService` now applies settings updates (e.g. via
  `STTUpdateSettingsFrame`) using a graceful reconnect instead of a hard
  disconnect/reconnect, preserving the service's reconnect retry behavior.
  (PR [#4521](https://github.com/pipecat-ai/pipecat/pull/4521))

- Updated the default p99 TTFS latency values for Smallest AI, Mistral, and XAI
  STT so turn stop timing uses measured values instead of the conservative
  fallback.
  (PR [#4522](https://github.com/pipecat-ai/pipecat/pull/4522))

- Updated the development runner startup banner to show the prebuilt client URL
  once and list enabled or disabled transports with install hints.
  (PR [#4524](https://github.com/pipecat-ai/pipecat/pull/4524))

- Services and transports with missing optional dependencies now raise
  `ImportError` instead of a bare `Exception` when their module is imported
  without the required extra installed. The original `ModuleNotFoundError` is
  preserved as `__cause__`, so code that wraps these imports can now use
  `except ImportError:` cleanly instead of `except Exception:`.
  (PR [#4525](https://github.com/pipecat-ai/pipecat/pull/4525))

- Bumped `pipecat-ai-prebuilt` to 1.0.1 in the `runner` extra, updating the
  prebuilt client UI served by the development runner.
  (PR [#4531](https://github.com/pipecat-ai/pipecat/pull/4531))

- Replaced the `transformers.WhisperFeatureExtractor` dependency in
  `LocalSmartTurnAnalyzerV3` with a vendored numpy-only implementation,
  reducing peak RSS at import from ~566 MB to ~60 MB and cold-start time from
  ~5.0 s to ~0.3 s. Behavior is numerically equivalent (matches the reference
  numpy code path within 1e-5 absolute tolerance; ONNX model output is
  bit-identical on representative inputs).
    - Smart Turn v3 no longer imports `transformers` at module load.
    - Prepares the ground for making `transformers` an optional dependency in a
      future release.
    - The vendored STFT is vectorized via
  `numpy.lib.stride_tricks.sliding_window_view` + batched `np.fft.rfft`,
  cutting `_power_spectrogram` runtime by ~55% (~4.0 ms → ~1.8 ms per call on a
  typical 8-second segment at 16 kHz) while preserving the same parity
  tolerances against the reference implementation.
  (PR [#4536](https://github.com/pipecat-ai/pipecat/pull/4536))

- ⚠️ Renamed the RTVI UI Worker Protocol's vocabulary from the
  `pipecat-subagents` `task`/`agent` terms to Pipecat's native `job`/`worker`.
  This spans the wire messages (`ui-task` → `ui-job-group`, `ui-cancel-task` →
  `ui-cancel-job-group`), their envelope `kind`s and fields (`task_id` →
  `job_id`, `agents`/`agent_name` → `workers`/`worker_name`), the paired Python
  models/frames (`UITask*` → `UIJobGroup*`, `RTVIUITask*Frame` →
  `RTVIUIJobGroup*Frame`), and the `@pipecat-ai/client-js` / `client-react`
  APIs (`RTVIEvent.UITask` → `UIJobGroup`, `cancelUITask` → `cancelUIJobGroup`,
  `useUITasks` → `useUIJobGroups`, `UITasksProvider` → `UIJobGroupsProvider`).
  These primitives shipped in 1.2.0 but were never documented, so no real
  consumers are affected.
  (PR [#4540](https://github.com/pipecat-ai/pipecat/pull/4540))

- `transformers` is no longer a base dependency, so `pip install pipecat-ai` no
  longer pulls it in. This follows Smart Turn v3 dropping its `transformers`
  import; the only remaining users (the deprecated
  `LocalSmartTurnAnalyzerV2`/CoreML analyzers and the Moondream service)
  already require the `local-smart-turn` and `moondream` extras, which continue
  to install `transformers`.
  (PR [#4546](https://github.com/pipecat-ai/pipecat/pull/4546))

- Widened the `deepgram` extra to `deepgram-sdk>=6.1.1,<8` so installations can
  resolve to either deepgram-sdk 6.x or 7.x. `DeepgramSTTService` now handles
  the `agent_rest` keyword argument that deepgram-sdk 7.2.0 added to
  `DeepgramClientEnvironment`, so custom `base_url` configuration keeps working
  on both 6.x and 7.x.
  (PR [#4565](https://github.com/pipecat-ai/pipecat/pull/4565))

- Dropped the upper bound on the `websockets-base` extra (`websockets>=13.1`)
  so downstream deployments can resolve to websockets 16.x and beyond.
  Pipecat's `websockets` usage relies only on the modern `websockets.asyncio`
  API plus a handful of public symbols, all of which are retained in 16.x.
  (PR [#4565](https://github.com/pipecat-ai/pipecat/pull/4565))

- Changed the default voice for `GradiumTTSService` to `_6Aslh2DxfmnRLmP`.
  (PR [#4569](https://github.com/pipecat-ai/pipecat/pull/4569))

- `InworldRealtimeLLMService` now defaults the STT model to
  `inworld/inworld-stt-1`.
  (PR [#4573](https://github.com/pipecat-ai/pipecat/pull/4573))

### Deprecated

- `FrameProcessor.pipeline_task` is deprecated; read
  `FrameProcessor.pipeline_worker` instead. The old name still works but emits
  a `DeprecationWarning` and will be removed in a future release.
  (PR [#4493](https://github.com/pipecat-ai/pipecat/pull/4493))

- Passing a worker to `WorkerRunner.run()` is deprecated. Register the worker
  with `WorkerRunner.add_workers()` before calling `run()` instead. The
  `worker` argument still works but emits a `DeprecationWarning` and will be
  removed in a future release.
  (PR [#4493](https://github.com/pipecat-ai/pipecat/pull/4493))

- `PipelineTask`, `PipelineTaskParams`, and the `pipecat.pipeline.task` module
  have been renamed to `PipelineWorker`, `WorkerParams`, and
  `pipecat.pipeline.worker`. The old names still resolve (the module re-exports
  the new symbols) but constructing `PipelineTask` / `PipelineTaskParams` emits
  a `DeprecationWarning`; they will be removed in a future release.
  (PR [#4493](https://github.com/pipecat-ai/pipecat/pull/4493))

- `PipelineRunner` has been renamed to `WorkerRunner` and moved to
  `pipecat.workers.runner`, since the runner now runs workers (of which
  `PipelineWorker` is one kind), not just pipelines. Import `WorkerRunner` from
  `pipecat.workers.runner`. The old `pipecat.pipeline.runner` module still
  re-exports both names, and `PipelineRunner` still works as a subclass alias,
  but it emits a `DeprecationWarning` and will be removed in a future release.
  (PR [#4589](https://github.com/pipecat-ai/pipecat/pull/4589))

### Removed

- Removed the unsupported Georgian (`Language.KA`) language mapping from
  `SonioxSTTService`.
  (PR [#4521](https://github.com/pipecat-ai/pipecat/pull/4521))

### Fixed

- Fixed Azure TTS last word being missed by observers and RTVI UI. The
  completion signal was racing with word timestamp processing, causing the
  final word's `TTSTextFrame` to arrive after `TTSStoppedFrame`. Completion is
  now routed through the word boundary queue to ensure all words are processed
  before signaling stream end.
  (PR [#4306](https://github.com/pipecat-ai/pipecat/pull/4306))

- Fixed skipped TTS frames (e.g. code blocks filtered via
  `skip_aggregator_types`) being emitted to the assistant context immediately
  instead of waiting for preceding spoken frames to finish. They now hold their
  position in the frame sequence and are flushed only after all earlier spoken
  sentences are complete, keeping context ordering correct.
  (PR [#4380](https://github.com/pipecat-ai/pipecat/pull/4380))

- Fixed Cartesia word timestamps leaking SSML tag text (e.g. `<spell>`,
  `<emotion>`, `<break>`) into word entries. Tags are now stripped before
  processing, so word-to-text attribution remains accurate when SSML markup is
  present in the TTS input.
  (PR [#4380](https://github.com/pipecat-ai/pipecat/pull/4380))

- Fixed `BaseOutputTransport` reordering frames that share the same
  presentation timestamp. Frames with equal PTS values are now emitted in
  insertion order, preventing subtle audio/text sequencing bugs when multiple
  frames arrive at the same time.
  (PR [#4380](https://github.com/pipecat-ai/pipecat/pull/4380))

- Fixed `TTSTextFrame` entries losing their original text structure when word
  timestamps are enabled. Each `TTSTextFrame` now carries a `raw_text` field
  containing the corresponding span of the original LLM-produced text
  (including pattern delimiters such as `<card>4111 1111 1111 1111</card>`), so
  the assistant context receives properly-tagged content rather than the
  cleaned words returned by the TTS provider. Also handles words that straddle
  two sentence boundaries by splitting them and attributing each part to its
  correct source frame.
  (PR [#4380](https://github.com/pipecat-ai/pipecat/pull/4380))

- Fixed `PipelineTask.cancel()` hanging when cancellation is requested before
  the initial `StartFrame` reaches the pipeline sink.
  (PR [#4455](https://github.com/pipecat-ai/pipecat/pull/4455))

- Fixed `SmallWebRTCClient.read_audio_frame` and `read_video_frame`
  busy-looping on `MediaStreamError`. When a track raises `MediaStreamError`, the
  generator now clears the track reference (`_audio_input_track` /
  `_video_input_track` / `_screen_video_track`) so the loop parks on the
  existing `is None` gate instead of re-entering `recv()` at ~100 Hz on a
  permanently-failed track. Renegotiation still resumes seamlessly: when
  `_handle_client_connected` reassigns a fresh track, the loop picks up frames
  from the new track.
  (PR [#4491](https://github.com/pipecat-ai/pipecat/pull/4491))

- Fixed `ElevenLabsSTTService` crashing when `language` was passed as `None`.
  When `language` is not set, the service now lets ElevenLabs auto-detect the
  audio language.
  (PR [#4507](https://github.com/pipecat-ai/pipecat/pull/4507))

- Fixed `NvidiaSTTService` so unexpected gRPC stream drops reconnect cleanly
  using the active audio iterator, while service shutdown and cancellation
  still close that iterator and stop the streaming worker without leaving it
  stuck waiting for more audio.
  (PR [#4512](https://github.com/pipecat-ai/pipecat/pull/4512))

- Fixed websocket STT connection setup failures so services clear stale
  websocket state and emit non-fatal error frames, allowing `ServiceSwitcher`
  failover to keep agents running.
  (PR [#4514](https://github.com/pipecat-ai/pipecat/pull/4514))

- Fixed `ElevenLabsTTSService` and `ElevenLabsHttpTTSService` inserting
  unwanted spaces between words when synthesizing Chinese or Japanese. Word
  timestamps for these languages already include their own spacing, so they are
  now forwarded with `includes_inter_frame_spaces=True` to avoid double-spacing
  in transcripts and context.
  (PR [#4517](https://github.com/pipecat-ai/pipecat/pull/4517))

- Fixed the development runner so missing optional transport dependencies
  disable only their related routes instead of failing startup in all-transport
  mode.
  (PR [#4524](https://github.com/pipecat-ai/pipecat/pull/4524))

- Fixed a race in `ElevenLabsTTSService` where the periodic keepalive could be
  sent for a new turn's context before that context's `voice_settings`
  initialization message, causing ElevenLabs to close the WebSocket with a 1008
  policy violation (`voice_settings field must be provided in the first message
  ...`). The keepalive now only targets a context once its context-init has
  been sent.
  (PR [#4527](https://github.com/pipecat-ai/pipecat/pull/4527))

- Switched `BaseSmartTurn` from `time.time()` to `time.monotonic()` for its
  three internal interval-math sites (audio-buffer timestamps, speech-start
  tracking, and the pre-speech buffer-trim loop). Wall-clock time can step
  forward or backward when NTP adjusts the system clock, which would silently
  corrupt the buffer trim (prune everything / prune nothing) and the
  speech-window extraction. The corrected primitive is monotonic and matches
  the existing `time.perf_counter()` usage already in place for
  inference-latency metrics.
  (PR [#4542](https://github.com/pipecat-ai/pipecat/pull/4542))

- Fixed `SOXRAudioResampler` and `SOXRStreamAudioResampler` ignoring the
  configured quality setting. Both resamplers were hardcoded to `VHQ`, which
  meant `RNNoiseFilter`'s `resampler_quality` argument (defaulting to `QQ`
  for low-latency real-time use) had no effect. The resamplers now honor
  the configured quality, with `VHQ` retained as the default.
  (PR [#4551](https://github.com/pipecat-ai/pipecat/pull/4551))

- Fixed `GeminiLiveLLMService` (and `GeminiVertexLiveLLMService`) crashing with
  `'ContextWindowCompressionParams' object has no attribute 'get'` when
  `context_window_compression` was supplied through the `settings` API (e.g.
  `settings=GeminiLiveLLMService.Settings(context_window_compression=ContextWindowCompressionParams(...))`).
  The setting is now handled whether it arrives as a
  `ContextWindowCompressionParams` object or as a dict.
  (PR [#4563](https://github.com/pipecat-ai/pipecat/pull/4563))

- Fixed `AudioBufferProcessor` concatenating utterances separated by a silent
  gap. When no user audio arrives for more than 200 ms, silence proportional to
  the wall-clock gap is now inserted into the user buffer; the same fix is
  applied symmetrically to the bot buffer, so two bot utterances spoken seconds
  apart (e.g. progressive hold messages played while a slow function call runs)
  remain temporally separated in the recorded audio.
  (PR [#4567](https://github.com/pipecat-ai/pipecat/pull/4567))

- `InworldRealtimeLLMService` no longer logs `WARNING`s for unrecognized
  realtime server events (e.g. `response.output_text.done`); they are now
  logged at `DEBUG`.
  (PR [#4573](https://github.com/pipecat-ai/pipecat/pull/4573))

- Fixed a spurious `ttfs_p99_latency not set, using default 1.0s` warning
  emitted by turn-based STT services (`CartesiaTurnsSTTService`,
  `DeepgramFluxSTTService`) at pipeline start. These services have no
  meaningful "speech end → final transcript" interval to measure, because the
  server defines turn boundaries directly.
  (PR [#4585](https://github.com/pipecat-ai/pipecat/pull/4585))

### Performance

- `BaseSmartTurn` now stores raw `int16` PCM views in its audio buffer and
  defers the `float32` conversion to the once-per-turn segment extraction,
  eliminating ~50 per-frame numpy allocations per second per analyzer. Output
  is bit-identical to the previous per-frame conversion path because `int16 →
  float32 / 32768.0` distributes over concatenation; subclasses
  (`LocalSmartTurnAnalyzerV3`, `LocalCoreMLSmartTurnAnalyzer`,
  `HttpSmartTurnAnalyzer`) all receive the same float32 `audio_array` they did
  before. Also removes a spurious `np.frombuffer(audio_int16, dtype=np.int16)`
  re-wrap that was a no-op view-of-a-view of already-int16 data.
  (PR [#4542](https://github.com/pipecat-ai/pipecat/pull/4542))

- Reduced the `soxr` resampling quality preset in `LocalSmartTurnAnalyzerV3`
  from `VHQ` (~26-tap polyphase) to `HQ` (~16-tap), cutting resample CPU time
  by 30–50% on non-16 kHz audio sources (~3–10 ms saved per turn at 24/48 kHz).
  Pipelines already delivering 16 kHz audio are unaffected — the existing
  `actual_rate == _MODEL_SAMPLE_RATE` fast path skips resampling entirely. The
  two quality presets differ in filter length, not cutoff or interpolation
  semantics; on a Whisper-style log-mel feature representation the audible
  difference sits well below the mel filterbank's quantization noise floor, so
  model predictions are unchanged on representative inputs.
  (PR [#4542](https://github.com/pipecat-ai/pipecat/pull/4542))

## [1.2.1] - 2026-05-15

### Changed

- Changed the default WebSocket endpoints for `GradiumSTTService` and
  `GradiumTTSService` to the region-neutral
  `wss://api.gradium.ai/api/speech/asr` and
  `wss://api.gradium.ai/api/speech/tts`. Gradium now automatically routes
  traffic to the nearest endpoint. Override the url to pin to a specific
  region.
  (PR [#4500](https://github.com/pipecat-ai/pipecat/pull/4500))

### Fixed

- Fixed bot hangs when `filter_incomplete_user_turns` was enabled and the LLM
  responded by calling a tool. The user turn never finalized, so the assistant
  aggregator gated the tool-result context push and the LLM continuation never
  ran. Tool calls now finalize the turn the moment they start, before the
  function dispatches.
  (PR [#4501](https://github.com/pipecat-ai/pipecat/pull/4501))

## [1.2.0] - 2026-05-14

### Added

- Added a `session_id` field to `RunnerArguments` so bots can log or trace a
  per-session identifier in local development the same way they can in Pipecat
  Cloud. The development runner now mints a UUID at every construction site,
  and paths that already returned a `sessionId` to the caller (Daily `/start`,
  dial-in webhook) share that same UUID with the runner args instead of
  generating two. The SmallWebRTC `/api/offer` endpoint also accepts an
  optional `session_id` query parameter so the `/sessions/{session_id}/...`
  proxy can thread it through.
  (PR [#4385](https://github.com/pipecat-ai/pipecat/pull/4385))

- Added a `max_buffer_delay_ms` constructor argument to `CartesiaTTSService`
  for controlling Cartesia's server-side text buffering. When unset, Pipecat
  picks a sensible default based on `text_aggregation_mode`: `0` in `SENTENCE`
  mode (custom buffering — avoids stacking client-side aggregation on top of
  Cartesia's default 3000ms server buffer) and unset in `TOKEN` mode
  (Cartesia's managed buffering applies). Pass an explicit value (0–5000ms) to
  override.
  (PR [#4390](https://github.com/pipecat-ai/pipecat/pull/4390))

- Added a `mip_opt_out` constructor argument to `DeepgramTTSService` and
  `DeepgramHttpTTSService` so callers can opt out of the Deepgram Model
  Improvement Program. When set, the value is forwarded to Deepgram as a query
  parameter on the speak request. Defaults to `None`, which preserves the
  existing behavior. See https://dpgr.am/deepgram-mip for pricing implications
  before enabling.
  (PR [#4400](https://github.com/pipecat-ai/pipecat/pull/4400))

- Added an opt-in `add_tool_change_messages` flag to the LLM aggregators (set
  via `LLMContextAggregatorPair(..., add_tool_change_messages=True)`) that
  appends a developer-role message to the context whenever `LLMSetToolsFrame`
  changes the set of advertised standard tools. Helps the LLM stay coherent
  across mid-conversation tool changes, mitigating several flavors of
  tool-call-related hallucination: calling tools that have been removed,
  avoiding tools that have been re-added, and hallucinating output (made-up
  answers or tool-call-shaped non-tool-calls) when tools are unavailable.
  (PR [#4404](https://github.com/pipecat-ai/pipecat/pull/4404))

- Added `deferred(strategy)` and `DeferredUserTurnStopStrategy` in
  `pipecat.turns.user_stop`. Wraps a stop strategy so it fires only the
  inference-triggered event and suppresses `on_user_turn_stopped`, leaving
  finalization to another strategy in the chain such as
  `LLMTurnCompletionUserTurnStopStrategy`.
  (PR [#4405](https://github.com/pipecat-ai/pipecat/pull/4405))

- Added `ExternalUserTurnCompletionStopStrategy` in `pipecat.turns.user_stop` —
  a generic stop strategy that finalizes the user turn whenever a
  `UserTurnInferenceCompletedFrame` arrives, regardless of which component
  produced it. `LLMTurnCompletionUserTurnStopStrategy` now extends this base;
  future producers (Flux, custom end-of-turn classifiers, etc.) can use the
  base directly or subclass it to add producer-specific setup.
  (PR [#4405](https://github.com/pipecat-ai/pipecat/pull/4405))

- Added `on_user_turn_inference_triggered`, a new event on the user turn
  controller, processor, aggregator and stop strategies that fires when a
  strategy has enough signal to start LLM inference. By default it fires
  together with `on_user_turn_stopped`; a gating strategy can fire only the
  inference-triggered event and defer finalization to a peer.
  (PR [#4405](https://github.com/pipecat-ai/pipecat/pull/4405))

- Added `FilterIncompleteUserTurnStrategies` in
  `pipecat.turns.user_turn_strategies` — a `UserTurnStrategies` specialization
  that wraps the detector chain with `deferred(...)` and appends
  `LLMTurnCompletionUserTurnStopStrategy` as the finalizer. Common case:
  `user_turn_strategies=FilterIncompleteUserTurnStrategies()`. Pass
  `config=UserTurnCompletionConfig(...)` to customize timeouts and prompts.
  (PR [#4405](https://github.com/pipecat-ai/pipecat/pull/4405))

- Added `LLMTurnCompletionUserTurnStopStrategy` in `pipecat.turns.user_stop`.
  When installed, the strategy gates `on_user_turn_stopped` on a
  `UserTurnInferenceCompletedFrame` (a new fieldless system frame emitted by
  any component that can judge turn completeness — e.g. the
  `UserTurnCompletionLLMServiceMixin` on `✓`). A `finalization_timeout`
  provides a safety net if no completion frame ever arrives.
  (PR [#4405](https://github.com/pipecat-ai/pipecat/pull/4405))

- Added first-class RTVI support for the UI Agent Protocol:
    - Adds `ui-event`, `ui-snapshot`, and `ui-cancel-task` client-to-server
  messages, plus `ui-command` and `ui-task` server-to-client messages, with
  paired `*Data` / `*Message` pydantic models.
    - Adds built-in command payload models for `Toast`, `Navigate`, `ScrollTo`,
  `Highlight`, `Focus`, `Click`, `SetInputValue`, and `SelectText`; matching
  default handlers live in `@pipecat-ai/client-react`.
    - Adds `RTVIProcessor.on_ui_message` for inbound `ui-event`, `ui-snapshot`,
  and `ui-cancel-task` messages.
    - Adds five UI pipeline frames, mirroring the `client-message`
  frame-and-event pattern: downstream code pushes `RTVIUICommandFrame` /
  `RTVIUITaskFrame` for the observer to wrap into outbound `UICommandMessage` /
  `UITaskMessage` envelopes, while the processor pushes inbound
  `RTVIUIEventFrame`, `RTVIUISnapshotFrame`, and `RTVIUICancelTaskFrame`
  alongside `on_ui_message`.
    - Bumps the RTVI `PROTOCOL_VERSION` from `1.2.0` to `1.3.0`.
  (PR [#4407](https://github.com/pipecat-ai/pipecat/pull/4407))

- AWS Transcribe STT, Polly TTS, Bedrock LLM, and the Bedrock AgentCore
  processor now resolve credentials via the standard boto3 provider chain (EC2
  instance profiles, EKS pod roles / IRSA, ECS task roles, SSO,
  `~/.aws/credentials`) when explicit credentials and `AWS_*` environment
  variables are absent. Services running with IAM roles no longer need to
  export static credentials.
  (PR [#4416](https://github.com/pipecat-ai/pipecat/pull/4416))

- Added `keyterms` support to ElevenLabs STT services so Scribe V2 callers can
  bias transcription for both file-based and realtime transcription.
  (PR [#4426](https://github.com/pipecat-ai/pipecat/pull/4426))

- Added `watchdog_min_timeout` parameter to `DeepgramFluxSTT` and
  `DeepgramFluxSageMakerSTT` (default `0.5` seconds) to control the minimum
  silence duration before the watchdog sends a silence packet to prevent
  dangling turns. The actual threshold is `max(chunk_duration * 2,
  watchdog_min_timeout)`, so it also adapts automatically to the audio chunk
  size in use.
  (PR [#4430](https://github.com/pipecat-ai/pipecat/pull/4430))

- Added `cancel_on_interruption=False` support for `GeminiLiveLLMService` on
  models that support Gemini's NON_BLOCKING tool mechanism (currently Gemini
  2.x); the conversation now continues while the tool runs. On models that
  don't yet support NON_BLOCKING (Gemini 3.x), the service surfaces a one-time
  warning explaining the limitation. (Note: an intermittent 1008 error can
  occasionally fire on Gemini 2.5 during long-running tool calls; we
  auto-reconnect.)
  (PR [#4448](https://github.com/pipecat-ai/pipecat/pull/4448))

- Added `NvidiaSageMakerWebsocketSTTService` for streaming speech recognition
  using NVIDIA Nemotron ASR via an AWS SageMaker bidirectional-stream endpoint.
  Produces `InterimTranscriptionFrame` and `TranscriptionFrame` frames, is
  VAD-aware, and automatically reconnects on error.
  (PR [#4464](https://github.com/pipecat-ai/pipecat/pull/4464))

- Added NVIDIA Magpie TTS services via AWS SageMaker:
  `NvidiaSageMakerHTTPTTSService` (single HTTP invocation, streams raw PCM
  back) and `NvidiaSageMakerWebsocketTTSService` (persistent HTTP/2 bidi-stream
  with full interruption support via `InterruptibleTTSService`).
  (PR [#4464](https://github.com/pipecat-ai/pipecat/pull/4464))

- Added support for `reasoning` configuration on `OpenAIRealtimeLLMService`,
  for use with reasoning-capable Realtime models such as `gpt-realtime-2`.
  (PR [#4470](https://github.com/pipecat-ai/pipecat/pull/4470))

- Inworld TTS updates:
    - Added `delivery_mode` setting (`STABLE`/`BALANCED`/`CREATIVE`) to
  `InworldTTSService` and `InworldHttpTTSService`, enabling the
  stability-vs-creativity tradeoff in `inworld-tts-2`.
    - Added language support to `InworldTTSService` and
  `InworldHttpTTSService`. The `language` setting is now forwarded to the API,
  and a new `language_to_inworld_language()` helper normalizes Pipecat
  `Language` enums to Inworld's BCP-47 locale tags.
  (PR [#4473](https://github.com/pipecat-ai/pipecat/pull/4473))

### Changed

- Updated the default `SonioxTTSService` model from `tts-rt-v1-preview` to the
  generally available `tts-rt-v1`.
  (PR [#4386](https://github.com/pipecat-ai/pipecat/pull/4386))

- Default `cartesia_version` for `CartesiaTTSService` bumped from `2025-04-16`
  to `2026-03-01`, matching `CartesiaHttpTTSService` and unlocking the
  `use_normalized_timestamps` and `max_buffer_delay_ms` fields.
  (PR [#4390](https://github.com/pipecat-ai/pipecat/pull/4390))

- ⚠️ `CartesiaTTSService` now sends `use_normalized_timestamps: true` instead
  of the deprecated `use_original_timestamps` field. Word timestamps now
  reflect what was actually spoken (post text-normalization and
  pronunciation-dictionary substitution), matching the convention Pipecat uses
  for ElevenLabs. This is a behavior change for `sonic-3` users, who were
  previously receiving timestamps tied to the input transcript.
  (PR [#4390](https://github.com/pipecat-ai/pipecat/pull/4390))

- Broadened `tool_resources` to `app_resources` for easy access not just in
  tool handlers but in other places like custom `FrameProcessor`s. Three
  changes: a rename (`tool_resources` → `app_resources`), a new `app_resources`
  property on `PipelineTask`, and a new `pipeline_task` property on
  `FrameProcessor`. Tool handlers now read `params.app_resources`; custom
  processors read `self.pipeline_task.app_resources`. The previous
  `tool_resources` aliases (on `PipelineTask`, `FunctionCallParams`, and
  `FrameProcessorSetup`) keep working but are deprecated as of 1.2.0 and emit
  `DeprecationWarning`s.
  (PR [#4395](https://github.com/pipecat-ai/pipecat/pull/4395))

- Lowered the per-message log in
  `SmallWebRTCInputTransport._handle_app_message` from `debug` to `trace`. App
  messages can be high-frequency and were noisy at debug level; set the loguru
  level to `TRACE` to see them again.
  (PR [#4397](https://github.com/pipecat-ai/pipecat/pull/4397))

- Changed the default model for `GrokRealtimeLLMService` to
  `grok-voice-think-fast-1.0`, xAI's recommended Voice Agent model. The
  previous default of `grok-voice-fast-1.0` has been deprecated by xAI and is
  being removed.
  (PR [#4401](https://github.com/pipecat-ai/pipecat/pull/4401))

- Changed the default Inworld TTS model from `inworld-tts-1.5-max` to
  `inworld-tts-2` (Realtime TTS-2) across `InworldHttpTTSService`,
  `InworldTTSService`, and the `InworldRealtimeLLMService` cascade. Existing
  users can pin the prior model explicitly via the `model`/`tts_model`
  argument; both `inworld-tts-1.5-max` and `inworld-tts-1.5-mini` remain valid
  model IDs.
  (PR [#4422](https://github.com/pipecat-ai/pipecat/pull/4422))

- Changed the default model for `GrokLLMService` from `grok-3` to
  `grok-4.20-non-reasoning`. xAI is retiring `grok-3` on May 15, 2026.
  (PR [#4429](https://github.com/pipecat-ai/pipecat/pull/4429))

- `DeepgramFluxSTT` watchdog silence threshold is now dynamic:
  `max(chunk_duration * 2, watchdog_min_timeout)` instead of a fixed 500 ms.
  This prevents false silence injections when large audio chunks are sent at
  lower frequency.
  (PR [#4430](https://github.com/pipecat-ai/pipecat/pull/4430))

- `ElevenLabsTTSService` now sends `close_context` to the server as soon as the
  turn is complete (on `on_turn_context_completed`) rather than waiting until
  all audio has finished playing back. The `isFinal` message from ElevenLabs is
  now used to signal `TTSStoppedFrame` and clean up the audio context,
  improving turn transition timing.
  (PR [#4433](https://github.com/pipecat-ai/pipecat/pull/4433))

- Updated `InworldHttpTTSService` and `InworldTTSService` to use PCM audio
  encoding by default, which returns audio bytes without headers.
  (PR [#4446](https://github.com/pipecat-ai/pipecat/pull/4446))

- Moved `create_task`, `cancel_task`, the `task_manager` property, and
  `setup(task_manager)` up from `FrameProcessor` to `BaseObject`. Custom
  `BaseObject` subclasses (turn strategies, controllers, etc.) now inherit
  these methods directly instead of reimplementing the task manager wiring.
  Owners propagate the task manager to their child `BaseObject`s via `await
  child.setup(task_manager)`.
  (PR [#4449](https://github.com/pipecat-ai/pipecat/pull/4449))

- Changed the default OpenAI Realtime input audio transcription model from
  `gpt-4o-transcribe` to `gpt-realtime-whisper` for both
  `OpenAIRealtimeSTTService` and `OpenAIRealtimeLLMService`. The new model does
  not accept the `prompt` parameter; if a prompt is supplied alongside
  `gpt-realtime-whisper`, it is dropped automatically and a warning is logged.
  To keep using prompt hints, explicitly pin `model="gpt-4o-transcribe"` (or
  `"gpt-4o-mini-transcribe"`).
  (PR [#4450](https://github.com/pipecat-ai/pipecat/pull/4450))

- Updated the default model for `CartesiaTTSService` and
  `CartesiaHttpTTSService` from `sonic-3` to `sonic-3.5`.
  (PR [#4462](https://github.com/pipecat-ai/pipecat/pull/4462))

- Changed the default model for `OpenAIRealtimeLLMService` from
  `gpt-realtime-1.5` to `gpt-realtime-2`.
  (PR [#4472](https://github.com/pipecat-ai/pipecat/pull/4472))

### Deprecated

- Deprecated `LLMUserAggregatorParams.filter_incomplete_user_turns`. Use
  `user_turn_strategies=FilterIncompleteUserTurnStrategies()` (or add
  `LLMTurnCompletionUserTurnStopStrategy` to a custom
  `user_turn_strategies.stop`) instead. Setting the legacy flag still works for
  one release: the aggregator emits a `DeprecationWarning` and rewires the
  strategies as if you had passed `FilterIncompleteUserTurnStrategies`
  directly.
  (PR [#4405](https://github.com/pipecat-ai/pipecat/pull/4405))

- Deprecated `ResampyResampler` in favor of `SOXRAudioResampler` (or the
  `create_file_resampler()` / `create_stream_resampler()` factories).
  Instantiating `ResampyResampler` now emits a `DeprecationWarning`. The class
  will be removed in Pipecat 2.0 along with the default `resampy` and `numba`
  dependencies.
  (PR [#4428](https://github.com/pipecat-ai/pipecat/pull/4428))

### Fixed

- Fixed `CartesiaTTSService` surfacing `flush_done` messages from Cartesia as
  `ErrorFrame`s. The latest API emits a `flush_done` per transcript when
  server-side buffering is disabled; Pipecat now consumes them silently since
  each turn already has its own `context_id`.
  (PR [#4390](https://github.com/pipecat-ai/pipecat/pull/4390))

- Fixed Cartesia tag helpers (`SPELL`, `EMOTION_TAG`, `PAUSE_TAG`,
  `VOLUME_TAG`, `SPEED_TAG`) raising `TypeError` when called on an instance
  (e.g. `tts.SPELL("hi")`). They're now `@staticmethod` and callable from both
  the class and an instance.
  (PR [#4390](https://github.com/pipecat-ai/pipecat/pull/4390))

- Fixed `CartesiaHttpTTSService` pushing two `ErrorFrame`s on a non-200
  response — one with the API's error text and a second, less informative
  "Unknown error" frame from the outer exception handler. It now pushes a
  single frame that includes the HTTP status code and returns cleanly.
  (PR [#4390](https://github.com/pipecat-ai/pipecat/pull/4390))

- Fixed an issue where `LocalSmartTurnAnalyzerV3` was imported unconditionally
  for user turn stop strategies. It is now only imported when
  `default_user_turn_stop_strategies()` is called. This improves startup time
  and removes the `transformers` "PyTorch/TensorFlow/Flax not found" warning
  when the default stop strategies are not used.
  (PR [#4393](https://github.com/pipecat-ai/pipecat/pull/4393))

- Fixed `GrokRealtimeLLMService` ignoring the configured model. The model was
  stored in `Settings` but never sent to xAI, so every session silently fell
  back to xAI's server-side default. The model is now passed via the `?model=`
  query parameter on the WebSocket URL as xAI's Voice Agent API requires.
  (PR [#4401](https://github.com/pipecat-ai/pipecat/pull/4401))

- Fixed `on_user_turn_stopped` firing prematurely when
  `filter_incomplete_user_turns` was enabled. The event now fires only after
  the LLM confirms the user turn is complete (`✓`); previously the smart-turn
  detector's tentative stop was bubbling up before the LLM had a chance to veto
  it, causing observers, transcript appenders and UI indicators to receive an
  early — and sometimes duplicated — signal.
  (PR [#4405](https://github.com/pipecat-ai/pipecat/pull/4405))

- Fixed `TTSSpeakFrame(append_to_context=True)` greetings sometimes splitting
  across two assistant messages in the LLM context and not surfacing in
  `on_assistant_turn_stopped`. The `LLMAssistantPushAggregationFrame` emitted
  at the end of a TTS context now carries a PTS just past the last word so it
  can't overtake clock-queued `TTSTextFrame`s in the transport's output, and
  `LLMAssistantAggregator` now triggers
  `on_assistant_turn_started`/`on_assistant_turn_stopped` when it receives the
  frame outside an LLM response cycle (restoring v0.0.104 behavior for greeting
  transcripts).
  (PR [#4414](https://github.com/pipecat-ai/pipecat/pull/4414))

- Fixed `ElevenLabsTTSService` and `ElevenLabsHttpTTSService` producing merged
  words (e.g. `bookLook`) when using Flash models. Flash often splits sentences
  mid-stream into alignment chunks that begin with a real inter-word space, but
  the previous fix unconditionally stripped that space from every chunk.
  Leading spaces are now stripped only on the first alignment chunk of an
  utterance, so subsequent chunks correctly flush partial words across
  boundaries.
  (PR [#4415](https://github.com/pipecat-ai/pipecat/pull/4415))

- Fixed AWS Polly TTS, Bedrock LLM, and the Bedrock AgentCore processor
  erroring out when only one of `AWS_ACCESS_KEY_ID` / `AWS_SECRET_ACCESS_KEY`
  was set in the environment. The half-populated kwargs are no longer forwarded
  to aioboto3; partial env-var configurations now fall through to the boto3
  credential chain like fully-unset configurations do.
  (PR [#4416](https://github.com/pipecat-ai/pipecat/pull/4416))

- Fixed `ElevenLabsTTSService` and `ElevenLabsHttpTTSService` writing
  romanized/normalized text to the LLM context. With non-Latin input (e.g.,
  Chinese), the assistant transcript was getting populated with pinyin (`Ni Hao
  !` instead of `你好！`), which then degraded subsequent LLM turns. The services
  now consume `alignment` by default and only switch to `normalizedAlignment` /
  `normalized_alignment` when `pronunciation_dictionary_locators` is configured
  (where `alignment` has overlapping restarts that produce duplicated/garbled
  words, per #4316). Both fields are read with preferred-with-fallback
  semantics since each is nullable per the API schema.
  (PR [#4424](https://github.com/pipecat-ai/pipecat/pull/4424))

- Fixed a deadlock in `TTSService` that could permanently stall pipeline
  processing when all three conditions occurred together:
  `pause_frame_processing=True`, an interruption arrived before any TTS audio
  was played, and an `UninterruptibleFrame` (e.g. `TTSUpdateSettingsFrame`,
  `FunctionCallResultFrame`) was in the processing queue at that moment. The
  process task would block on `__process_event.wait()` indefinitely because
  `BotStoppedSpeakingFrame` never arrives (no audio was played) and the
  interruption handler did not resume processing. Affects services using
  `pause_frame_processing=True` such as ElevenLabs, Rime, AsyncAI, Gradium, and
  ResembleAI.
  (PR [#4431](https://github.com/pipecat-ai/pipecat/pull/4431))

- Fixed interruptions being delayed when a slow non-uninterruptible frame was
  processing and an uninterruptible frame was waiting in the queue. The bot
  would stall until the slow frame finished instead of cancelling it
  immediately on interruption.
  (PR [#4434](https://github.com/pipecat-ai/pipecat/pull/4434))

- Fixed `TTSService` dropping uninterruptible frames (e.g.
  `FunctionCallResultFrame`) from its internal serialization queue when an
  interruption occurs. Previously, the queue was recreated on every
  interruption, silently discarding any queued frames. The queue is now reset
  instead of recreated, preserving uninterruptible frames so they are always
  delivered downstream.
  (PR [#4435](https://github.com/pipecat-ai/pipecat/pull/4435))

- Fixed a race condition in the Daily transport that caused `AttributeError:
  'NoneType' object has no attribute 'send_app_message'` when tearing down a
  pipeline. Both `DailyInputTransport` and `DailyOutputTransport` share the
  same `DailyTransportClient` and both call `cleanup()`, which was releasing
  the underlying `CallClient` on the first call — leaving the second caller
  with a `None` client.
  (PR [#4440](https://github.com/pipecat-ai/pipecat/pull/4440))

- Restored `cancel_on_interruption=False` support for `AWSNovaSonicLLMService`
  and `OpenAIRealtimeLLMService`. These services previously honored the flag by
  simply not cancelling in-flight function calls on interruption; the
  introduction of the new async-tool mechanism (which threads
  started/intermediate/final messages through the LLM context) broke that path
  because the realtime services didn't know how to interpret those messages.
  Note that new-style streamed intermediate results
  (`FunctionCallResultProperties(is_final=False)`) are not supported on these
  realtime services. Similar fixes for other impacted realtime services are
  forthcoming.
  (PR [#4441](https://github.com/pipecat-ai/pipecat/pull/4441))

- Fixed two misspelled Gemini TTS voice names in
  `GeminiTTSService.AVAILABLE_VOICES`.
  (PR [#4443](https://github.com/pipecat-ai/pipecat/pull/4443))

- Extended the `cancel_on_interruption=False` regression fix to
  `GrokRealtimeLLMService`, `AzureRealtimeLLMService`, and
  `UltravoxRealtimeLLMService`. Grok and Azure use the same approach as in
  #4441 (each service detects async-tool messages in the LLM context and routes
  the final result to its formal tool-result channel; Azure inherits
  transitively from `OpenAIRealtimeLLMService`). Ultravox needed a different
  approach because its API freezes the conversation between
  `client_tool_invocation` and the matching `client_tool_result` — for
  async-registered functions it now ships a placeholder `client_tool_result`
  immediately when the function is invoked (to unfreeze the conversation), then
  injects the real result as user-side text once the tool finishes. Streamed
  intermediate results (`FunctionCallResultProperties(is_final=False)`) are
  still not supported on any of these realtime services. `GeminiLiveLLMService`
  and `InworldRealtimeLLMService` are excluded for now: Gemini Live's
  async-tool path needs deeper investigation, and Inworld tool calling needs to
  be sorted out first.
  (PR [#4447](https://github.com/pipecat-ai/pipecat/pull/4447))

- Fixed `OpenAIRealtimeLLMService` handling of multi-output-item responses
  (observed with `gpt-realtime-2`). A single response can now contain more than
  one audio item, and the first item's `audio.done` may arrive after the second
  item's deltas have started. Deltas still arrive strictly in playback order,
  so we continue to forward them as received (matching OpenAI's reference
  implementation). The fix removes spurious warnings, ensures truncation always
  targets the latest audio item, and emits a single bracketing
  `TTSStartedFrame`/`TTSStoppedFrame` pair per assistant turn (the Stopped is
  now pushed on `response.done`).
  (PR [#4465](https://github.com/pipecat-ai/pipecat/pull/4465))

- Fixed missing `output` attribute on LLM OpenTelemetry spans when the LLM call
  is interrupted mid-stream.
  (PR [#4467](https://github.com/pipecat-ai/pipecat/pull/4467))

- Fixed incorrect `metrics.ttfb` on STT OpenTelemetry spans, and parented them
  to the current turn span.
  (PR [#4467](https://github.com/pipecat-ai/pipecat/pull/4467))

- Fixed incorrect `metrics.ttfb` on TTS OpenTelemetry spans for streaming
  services.
  (PR [#4467](https://github.com/pipecat-ai/pipecat/pull/4467))

- Extended the `cancel_on_interruption=False` regression fix to
  `InworldRealtimeLLMService`. Uses the same approach as in #4441 (the service
  detects async-tool messages in the LLM context and routes the final result to
  its formal tool-result channel). Note: as of this writing, Inworld Realtime
  doesn't appear to handle the resulting delayed tool result reliably — the
  routing is best-effort and the service surfaces a one-time warning when
  async-tool messages are seen. Streamed intermediate results
  (`FunctionCallResultProperties(is_final=False)`) are still not supported on
  this realtime service. (Inworld was excluded from #4447 pending resolution of
  an unrelated tool-calling issue, which turned out to be an account-level
  matter.)
  (PR [#4474](https://github.com/pipecat-ai/pipecat/pull/4474))

- Fixed Cartesia TTS Korean word timestamps to use normal spacing rules,
  preserving word boundaries and per-word timestamp alignment during downstream
  aggregation.
  (PR [#4475](https://github.com/pipecat-ai/pipecat/pull/4475))

- Fixed Cartesia TTS Chinese and Japanese timestamp grouping to preserve
  provider text spacing, avoiding artificial spaces when timestamp groups are
  reassembled downstream.
  (PR [#4475](https://github.com/pipecat-ai/pipecat/pull/4475))

- Fixed `SonioxSTTService` final transcription frames missing detected language
  metadata when Soniox returns token-level language annotations.
  (PR [#4482](https://github.com/pipecat-ai/pipecat/pull/4482))

- Fixed Soniox final transcription language detection to use the most common
  recognized token language, avoiding mislabeling an utterance when the last
  token is tagged with a different language.
  (PR [#4495](https://github.com/pipecat-ai/pipecat/pull/4495))

- Fixed dropped audio in streaming TTS services whose wire protocol doesn't
  echo `context_id` back on incoming audio (Sarvam, Smallest, Soniox, Inworld,
  and others). Previously, audio that arrived between contexts or at the very
  start of a turn was tagged with `context_id=None` and silently dropped with
  an "unable to append audio to context: no context ID provided" debug log.
  `TTSService.get_active_audio_context_id()` now falls back to the
  synthesis-side `_turn_context_id` when the playback cursor isn't set yet.
  (PR [#4497](https://github.com/pipecat-ai/pipecat/pull/4497))

### Security

- Fixed a path traversal issue in the development runner's
  `/files/{filename:path}` download endpoint. Previously, when the runner was
  started with `--folder`, a request like `/files/..%2F..%2Fetc%2Fpasswd` could
  escape the configured folder because `%2F`-encoded separators bypassed
  Starlette's path normalisation. The endpoint now resolves the joined path and
  rejects any filename that escapes the allowed base with a 403, and also
  returns 404 (instead of an implicit `null` 200) when `--folder` is unset.
  (PR [#4417](https://github.com/pipecat-ai/pipecat/pull/4417))

## [1.1.0] - 2026-04-27

### Added

- Added `MistralSTTService` for real-time speech-to-text using Mistral's
  Voxtral Realtime API (`voxtral-mini-transcribe-realtime-2602`). Supports
  streaming transcription with interim results, automatic language detection,
  and VAD-driven utterance lifecycle.
  (PR [#4253](https://github.com/pipecat-ai/pipecat/pull/4253))

- Added `buttons` field to `OutputDTMFFrame` and `OutputDTMFUrgentFrame` for
  sending multi-key DTMF sequences as a `list[KeypadEntry]`. Use
  `OutputDTMFFrame.from_string("123#")` (or the equivalent on
  `OutputDTMFUrgentFrame`) to build one from a dial string, and `to_string()`
  to convert back.
  (PR [#4313](https://github.com/pipecat-ai/pipecat/pull/4313))

- Added `DailyTransport.send_dtmf()` to expose the Daily call client's DTMF
  sending capability, enabling applications to send tones during a call (e.g.
  IVR navigation).
  (PR [#4313](https://github.com/pipecat-ai/pipecat/pull/4313))

- Added `DailyOutputDTMFFrame` and `DailyOutputDTMFUrgentFrame` frames. In
  addition to the inherited `buttons`, they accept `session_id`,
  `digit_duration_ms` and `method`, which are forwarded to Daily's `send_dtmf`
  as `sessionId`, `digitDurationMs` and `method`.
  (PR [#4313](https://github.com/pipecat-ai/pipecat/pull/4313))

- Added incremental `pyright` type checking. A `pyrightconfig.json` at the repo
  root uses `typeCheckingMode: "basic"` with an explicit `include` list of
  modules that pass cleanly (`clocks`, `metrics`, `transcriptions`, `frames`,
  `observers`, `extensions`, `turns`, `pipeline`, `runner`). Remaining modules
  will be added in subsequent PRs. CI enforces the checked set via `uv run
  pyright` in the format workflow.
  (PR [#4324](https://github.com/pipecat-ai/pipecat/pull/4324))

- Added multilingual support to `DeepgramFluxSTTService` via a new
  `language_hints: list[Language]` setting. Works with Deepgram's new
  `flux-general-multi` model to bias transcription across English, Spanish,
  French, German, Hindi, Russian, Portuguese, Japanese, Italian, and Dutch.
  Omit the hints to use auto-detection, or pass a subset to bias toward
  expected languages. Hints can be updated mid-stream via
  `STTUpdateSettingsFrame` (sent as a Deepgram `Configure` control message, no
  reconnect) to support detect-then-lock flows.
  (PR [#4326](https://github.com/pipecat-ai/pipecat/pull/4326))

- Added fine-grained server-side VAD tuning options to
  `SarvamSTTService.Settings` for the `saaras:v3` model, including speech
  thresholds, frame-count controls, pre-speech padding, interruption
  sensitivity, and initial-frame skipping.
  (PR [#4334](https://github.com/pipecat-ai/pipecat/pull/4334))

- Added `XAISTTService` for real-time speech-to-text using xAI's voice STT
  WebSocket API (`wss://api.x.ai/v1/stt`). Streams raw audio (PCM, µ-law, or
  A-law) and emits interim and final transcription frames driven by the
  server's `is_final` / `speech_final` flags. Settings expose
  `interim_results`, `endpointing`, `language`, `multichannel`, `channels`, and
  `diarize`. Requires the `xai` optional extra (`pip install
  "pipecat-ai[xai]"`).
  (PR [#4340](https://github.com/pipecat-ai/pipecat/pull/4340))

- Added `XAITTSService` for streaming text-to-speech using xAI's WebSocket TTS
  endpoint (`wss://api.x.ai/v1/tts`). Streams `text.delta` chunks up and base64
  `audio.delta` chunks down on the same connection so audio begins flowing
  before the full utterance finishes synthesizing; complements the batch-HTTP
  `XAIHttpTTSService`. Defaults to raw PCM output so `TTSAudioRawFrame` needs
  no decoding. The `xai` optional extra now pulls in
  `pipecat-ai[websockets-base]`.
  (PR [#4341](https://github.com/pipecat-ai/pipecat/pull/4341))

- Added `SonioxTTSService`, a real-time WebSocket TTS service that streams text
  in and audio out over a persistent connection. Install with `pip install
  "pipecat-ai[soniox]"`.
  (PR [#4360](https://github.com/pipecat-ai/pipecat/pull/4360))

- Added support for Daily's built-in `screenVideo` destination in
  `DailyTransport`. When `"screenVideo"` is included in
  `video_out_destinations` transport parameter, a dedicated screen video track
  is created at join time and frames with `transport_destination="screenVideo"`
  are routed to it.

    ```python
    params = DailyParams(
        video_out_enabled=True,
        video_out_is_live=True,
        video_out_width=1280,
        video_out_height=720,
        video_out_destinations=["screenVideo"],
    )

    ...

    frame = OutputImageRawFrame(...)
    frame.transport_destination = "screenVideo"
    ```
  (PR [#4370](https://github.com/pipecat-ai/pipecat/pull/4370))

- Added `camera_out_send_settings` to `DailyParams`. This dict is passed
  verbatim to the Daily client's camera publishing settings, allowing
  applications to fully control encoding, codec, bitrate, and framerate.

    ```python
    params = DailyParams(
        camera_out_send_settings={
            "maxQuality": "high",
            "encodings": {"high": {"maxBitrate": 2_000_000, "maxFramerate": 30}},
        },
    )
    ```
  (PR [#4370](https://github.com/pipecat-ai/pipecat/pull/4370))

- Added `tool_resources` to `PipelineTask` and `FunctionCallParams`. Pass an
  application-defined object (DB handles, clients, state, etc.) to
  `PipelineTask(..., tool_resources=...)` and access it from any tool handler
  via `params.tool_resources`. Passed by reference; the caller retains their
  handle and can read mutations after the task finishes. Resolves #4256.
  (PR [#4371](https://github.com/pipecat-ai/pipecat/pull/4371))

### Changed

- Updated NVIDIA STT services to align with Nemotron Speech defaults and
  configuration: `api_key` is now optional for local deployments, additional
  recognition settings are available (including alternatives, word offsets, and
  diarization), and streaming/segmented docs now reflect Nemotron Speech APIs.
  - NVIDIA streaming STT now sets `TranscriptionFrame.finalized=True` when the
  provider marks a result as final, and preserves `language` on both
  `TranscriptionFrame` and `InterimTranscriptionFrame`.
  (PR [#4269](https://github.com/pipecat-ai/pipecat/pull/4269))

- Updated `NvidiaLLMService` to emit model reasoning as `LLMThought*Frame`s
  (from both `reasoning_content` and `<think>...</think>` output), avoid mixing
  reasoning text into normal assistant content, and allow keyless local NIM
  endpoints while warning when the cloud endpoint is used without an API key.
  (PR [#4270](https://github.com/pipecat-ai/pipecat/pull/4270))

- STT services now reconnect safely when settings change: reconnection is
  deferred until the current user turn ends (i.e., until
  `UserStoppedSpeakingFrame` is received) rather than interrupting an active
  speech session. Audio frames received while the reconnect is in progress are
  buffered and replayed once the new connection is ready. `CartesiaSTTService`
  and `DeepgramSTTService` both use this new behavior.
  (PR [#4311](https://github.com/pipecat-ai/pipecat/pull/4311))

- Reduced debug log noise for LLM services. The system instruction is now
  logged once when composed (e.g. when turn completion is enabled) instead of
  on every LLM call. Per-call logs now show only the conversation messages,
  consistent across Google, Anthropic, AWS, and OpenAI services.
  (PR [#4314](https://github.com/pipecat-ai/pipecat/pull/4314))

- `LiveKitRunnerArguments.token` is now a required `str` (previously `str |
  None` with a default of `None`). LiveKit requires a token to join a room, so
  the type now reflects reality. This only affects custom runners that
  construct `LiveKitRunnerArguments` directly; code consuming the argument from
  the standard runner is unaffected.
  (PR [#4324](https://github.com/pipecat-ai/pipecat/pull/4324))

- `TranscriptionFrame.language` and `InterimTranscriptionFrame.language`
  emitted by `DeepgramFluxSTTService` now reflect the language Deepgram
  detected for each turn (read from the `languages` field on Flux's `TurnInfo`
  event). On `flux-general-multi` this gives per-turn accuracy for downstream
  consumers (e.g. TTS voice selection). `flux-general-en` continues to emit
  `Language.EN`.
  (PR [#4326](https://github.com/pipecat-ai/pipecat/pull/4326))

- Added `includes_inter_frame_spaces` parameter to
  `TTSService.add_word_timestamps` and `_add_word_timestamps` (default `None`).
  When `True`, downstream consumers will not inject additional spaces between
  tokens; `None` leaves each frame's own default unchanged.
  - `InworldTTSService` now passes `includes_inter_frame_spaces=True` when
  reporting word timestamps, since Inworld tokens already include inter-word
  spacing.
  (PR [#4330](https://github.com/pipecat-ai/pipecat/pull/4330))

- `SarvamSTTService` now uses `saaras:v3` as its default model instead of
  `saarika:v2.5`. Applications that relied on the previous default should set
  `settings=SarvamSTTService.Settings(model="saarika:v2.5")` explicitly.
  (PR [#4334](https://github.com/pipecat-ai/pipecat/pull/4334))

- `SpeechTimeoutUserTurnStopStrategy` now waits only `user_speech_timeout` when
  a transcript arrives without a VAD stop event, rather than
  `max(ttfs_p99_latency, user_speech_timeout)`. If you had `ttfs_p99_latency >
  user_speech_timeout`, turn detection in that path is slightly faster than
  before.
  (PR [#4337](https://github.com/pipecat-ai/pipecat/pull/4337))

- If you use an STT service that emits finalized transcripts (Speechmatics,
  Soniox, Deepgram Flux, AssemblyAI) with `SpeechTimeoutUserTurnStopStrategy`,
  user turns now end as soon as `user_speech_timeout` elapses after VAD stop.
  Previously the strategy also waited for the STT P99 latency
  (`ttfs_p99_latency`) even when the transcript was already marked final.
  `user_speech_timeout` is still honored as a floor — STT finalization never
  shortens it.
  (PR [#4337](https://github.com/pipecat-ai/pipecat/pull/4337))

- ⚠️ `PlivoFrameSerializer` and `TelnyxFrameSerializer` now raise `ValueError`
  at construction when `auto_hang_up=True` (the default) but required
  credentials are missing, matching `TwilioFrameSerializer`. Previously they
  constructed successfully and the hangup failed silently at call-end, leaving
  phantom billable sessions on the provider. If you relied on the old silent
  behavior, pass `auto_hang_up=False` explicitly or provide the credentials.
  The specific fields checked are `call_id`/`auth_id`/`auth_token` for Plivo
  and `call_control_id`/`api_key` for Telnyx.
  (PR [#4349](https://github.com/pipecat-ai/pipecat/pull/4349))

- `ToolsSchema(standard_tools=...)` now accepts any `Sequence[FunctionSchema |
  DirectFunction]` rather than requiring an exact `list` of the union. Callers
  can pass a narrower `list[FunctionSchema]` (or any other `Sequence`) without
  the type checker complaining about list invariance.
  (PR [#4352](https://github.com/pipecat-ai/pipecat/pull/4352))

- Updated `aic-sdk` dependency to `~=2.2.0`. The `AIC_LICENSE_KEY` environment
  variable replaces the previous `AICOUSTICS_LICENSE_KEY`.
  (PR [#4362](https://github.com/pipecat-ai/pipecat/pull/4362))

- Loosened the `protobuf` dependency to `>=5.29.6,<7`, so projects pinned to
  protobuf 5.x can install `pipecat-ai` again. The previous `>=6.31.1,<7` pin
  (introduced in 1.0.8 alongside the `nvidia-riva-client 2.25.1` upgrade)
  silently blocked any environment whose dependency graph already constrained
  protobuf to the 5.x line. The bundled `frames_pb2.py` is now compiled with
  protoc 5.x so it imports cleanly on both 5.x and 6.x runtimes.

  Installing the `nvidia` extra still pulls protobuf 6.x: `nvidia-riva-client
  2.25.1` ships gencode that requires a 6.x runtime, so `pipecat-ai[nvidia]`
  now declares `protobuf>=6.31.1,<7` explicitly to cover an upstream packaging
  gap (https://github.com/nvidia-riva/python-clients/issues/172).
  (PR [#4372](https://github.com/pipecat-ai/pipecat/pull/4372))

- Daily rooms created by the development runner (`pipecat.runner.run`) now
  expire after 4 hours with `eject_at_room_exp=True`, mirroring Pipecat Cloud's
  max session limit. Previously, runner-created rooms inherited a 2-hour
  expiration on the default code paths and had no expiration at all when
  callers posted partial `dailyRoomProperties` (e.g. `{"start_video_off":
  true}`) to `/start`, causing rooms to accumulate indefinitely. Explicit `exp`
  and `eject_at_room_exp` values in `dailyRoomProperties` are still respected.
  (PR [#4374](https://github.com/pipecat-ai/pipecat/pull/4374))

- Updated `daily-python` dependency to `~=0.28.0`.
  (PR [#4379](https://github.com/pipecat-ai/pipecat/pull/4379))

### Deprecated

- Deprecated `TransportParams.video_out_bitrate` for the Daily transport. Use
  `DailyParams.camera_out_send_settings` instead to configure camera publishing
  encodings (bitrate, framerate, codec, etc.).
  (PR [#4370](https://github.com/pipecat-ai/pipecat/pull/4370))

### Fixed

- Fixed missing tool handlers so unregistered tool calls fail with a normal
  final tool result instead of leaving tool-call state hanging.
  (PR [#4301](https://github.com/pipecat-ai/pipecat/pull/4301))

- Fixed `pipecat-ai[tavus]` not installing the required `daily-python`
  dependency. Installing the `tavus` extra now correctly pulls in
  `pipecat-ai[daily]`.
  (PR [#4304](https://github.com/pipecat-ai/pipecat/pull/4304))

- Fixed audio loss and potential errors when STT settings were updated
  mid-speech. Previously, `CartesiaSTTService` and `DeepgramSTTService` would
  immediately disconnect and reconnect when settings changed, dropping any
  in-flight audio. Reconnection is now deferred until the user stops speaking,
  and audio arriving during the reconnect window is buffered and replayed.
(PR [#4311](https://github.com/pipecat-ai/pipecat/pull/4311))

- Fixed `SmallestTTSService` WebSocket endpoint URL to match Smallest AI v4.0.0
  API (`wss://waves-api.smallest.ai` → `wss://api.smallest.ai`) and restored
  keepalive using a silent space message instead of the unsupported flush
  command.
  (PR [#4320](https://github.com/pipecat-ai/pipecat/pull/4320))

- Fixed whitespace handling in TTS token streaming mode. Inter-token whitespace
  (e.g., spaces between words) is now preserved for correct prosody, while
  leading whitespace before the first non-whitespace token is still stripped to
  avoid issues with TTS models that are sensitive to leading spaces.
  (PR [#4323](https://github.com/pipecat-ai/pipecat/pull/4323))

- Fixed `SentryMetrics` silently dropping `MetricsFrame`s from
  `stop_ttfb_metrics` and `stop_processing_metrics`. `SentryMetrics` called the
  base `FrameProcessorMetrics` implementation but discarded its return value,
  so `FrameProcessor` never pushed the `MetricsFrame` downstream. This
  prevented observers (e.g. `UserBotLatencyObserver`, `MetricsLogObserver`)
  from seeing TTFB and processing metrics for any service using
  `metrics=SentryMetrics()`. The metrics were still calculated and Sentry
  transactions still completed — only the downstream frame push was affected.
  (PR [#4325](https://github.com/pipecat-ai/pipecat/pull/4325))

- Fixed `ElevenLabsTTSService` and `ElevenLabsHttpTTSService` emitting word
  timestamps and `TTSTextFrame` content that matched the input text instead of
  the spoken audio when a pronunciation dictionary
  (`pronunciation_dictionary_locators`) or text normalization rewrote the
  input. Both services now consume ElevenLabs' normalized alignment, so
  downstream consumers (captions, transcripts, context aggregation) reflect
  what the listener actually hears.
  (PR [#4344](https://github.com/pipecat-ai/pipecat/pull/4344))

- Fixed a crash in `DeepgramSTTService` when an `STTUpdateSettingsFrame`
  arrived before the WebSocket handshake completed (for example, when pushing
  an update upstream on `StartFrame`). The settings-triggered reconnect
  cancelled the in-flight connection task before its keepalive task was
  created, causing an `UnboundLocalError: cannot access local variable
  'keepalive_task'` in the handler's `finally` block.
  (PR [#4347](https://github.com/pipecat-ai/pipecat/pull/4347))

- Fixed direct-function registration crashing for functions without a
  docstring. `DirectFunctionWrapper` passed `inspect.getdoc()`'s result to
  `docstring_parser.parse()`, which raises when the docstring is `None`.
  Functions now register cleanly whether or not they have a docstring; an empty
  docstring produces empty description and parameter metadata as expected.
  (PR [#4352](https://github.com/pipecat-ai/pipecat/pull/4352))

- Fixed `AssemblyAISTTService`, `CartesiaSTTService`, `GradiumSTTService`, and
  `SonioxSTTService` crashing the pipeline on transient WebSocket send
  failures. Each `run_stt` sent audio directly without catching errors, so a
  single network hiccup mid-stream raised an uncaught exception through
  `process_frame`. The guards now log a warning and let the connection-state
  check on the next call handle recovery, matching the pattern used by
  Deepgram, xAI, Azure, and other push-based STTs.
  (PR [#4352](https://github.com/pipecat-ai/pipecat/pull/4352))

- Fixed Gemini Live losing conversation history in the (rare) case of a
  WebSocket reconnect before any session resumption handle is received. When
  the session reconnects (e.g. on system instruction change), conversation
  history is now re-seeded into the new session before it is marked ready for
  input.
  (PR [#4355](https://github.com/pipecat-ai/pipecat/pull/4355))

- Fixed SmallWebRTC data channel silently stalling on networks with a 1280-byte
  MTU (IPv6, Tailscale overlays, many consumer VPNs). aiortc's default SCTP
  chunk size of 1200 bytes produces ~1305-byte UDP datagrams after headers,
  which the kernel rejects with EMSGSIZE; aiortc has no path-MTU discovery so
  it retransmits forever at the same oversized size. The chunk size is now
  clamped to 1100 bytes (~1205-byte datagrams, ~75 bytes of slack). Override
  with `PIPECAT_SCTP_MAX_CHUNK_SIZE` if your path MTU requires a different
  value.
  (PR [#4358](https://github.com/pipecat-ai/pipecat/pull/4358))

## [1.0.0] - 2026-04-14

Migration guide: https://docs.pipecat.ai/pipecat/migration/migration-1.0

### Added

- Updated LemonSlice transport:
    - Added `on_avatar_connected` and `on_avatar_disconnected` events triggered
      when the avatar joins and leaves the room.
    - Added `api_url` parameter to `LemonSliceNewSessionRequest` to allow
      overriding the LemonSlice API endpoint.
    - Added support for passing arbitrary named parameters to the LemonSlice
      API endpoint.
  (PR [#3995](https://github.com/pipecat-ai/pipecat/pull/3995))

- Added Inworld Realtime LLM service with WebSocket-based cascade STT/LLM/TTS,
  semantic VAD, function calling, and Router support.
  (PR [#4140](https://github.com/pipecat-ai/pipecat/pull/4140))

- ⚠️ Added WebSocket-based `OpenAIResponsesLLMService` as the new default for
  the OpenAI Responses API. It maintains a persistent connection to
  `wss://api.openai.com/v1/responses` and automatically uses
  `previous_response_id` to send only incremental context, falling back to full
  context on reconnection or cache miss. The previous HTTP-based implementation
  is now available as `OpenAIResponsesHttpLLMService`.
  (PR [#4141](https://github.com/pipecat-ai/pipecat/pull/4141))

- Added `group_parallel_tools` parameter to `LLMService` (default `True`). When
  `True`, all function calls from the same LLM response batch share a group ID
  and the LLM is triggered exactly once after the last call completes. Set to
  `False` to trigger inference independently for each function call result as
  it arrives.
  (PR [#4217](https://github.com/pipecat-ai/pipecat/pull/4217))

- Added async function call support to `register_function()` and
  `register_direct_function()` via `cancel_on_interruption=False`. When set to
  `False`, the LLM continues the conversation immediately without waiting for
  the function result. The result is injected back into the context as a
  `developer` message once available, triggering a new LLM inference at that
  point.
  (PR [#4217](https://github.com/pipecat-ai/pipecat/pull/4217))

- Added `enable_prompt_caching` setting to `AWSBedrockLLMService` for Bedrock
  ConverseStream prompt caching.
  (PR [#4219](https://github.com/pipecat-ai/pipecat/pull/4219))

- Added support for streaming intermediate results from async function calls.
  Call `result_callback` multiple times with
  `properties=FunctionCallResultProperties(is_final=False)` to push incremental
  updates, then call it once more (with `is_final=True`, the default) to
  deliver the final result. Only valid for functions registered with
  `cancel_on_interruption=False`.
  (PR [#4230](https://github.com/pipecat-ai/pipecat/pull/4230))

- Added `LLMMessagesTransformFrame` to facilitate programmatically editing
  context in a frame-based way.

  The previous approach required the caller to directly grab a reference to
  the context object, grab a "snapshot" of its messages _at that point in
  time_, transform the messages, and then push an `LLMMessagesUpdateFrame` with
  the transformed messages. This approach can lead to problems: what if there
  had already been a change to the context queued in the pipeline? The
  transformed messages would simply overwrite it without consideration.
  (PR [#4231](https://github.com/pipecat-ai/pipecat/pull/4231))

- The development runner now exports a module-level `app` FastAPI instance
  (`from pipecat.runner.run import app`) so you can register custom routes
  before calling `main()`.
  (PR [#4234](https://github.com/pipecat-ai/pipecat/pull/4234))

- `ToolsSchema` now accepts `custom_tools` for OpenAI LLM services
  (`OpenAILLMService`, `OpenAIResponsesLLMService`,
  `OpenAIResponsesHttpLLMService`, and `OpenAIRealtimeLLMService`), letting you
  pass provider-specific tools like `tool_search` alongside standard function
  tools.
  (PR [#4248](https://github.com/pipecat-ai/pipecat/pull/4248))

- Added enhancements to `NvidiaTTSService`:

  - Cross-sentence stitching: multiple sentences within an LLM turn are fed
    into a single `SynthesizeOnline` gRPC stream for seamless audio across
    sentence boundaries (requires Magpie TTS model v1.7.0+).
  - `custom_dictionary` and `encoding` parameters for IPA-based custom
    pronunciation and output audio encoding.
  - Metrics generation (`can_generate_metrics` returns true) and
    `stop_all_metrics()` when an audio context is interrupted.
  - gRPC error handling around synthesis config retrieval
    (`GetRivaSynthesisConfig`).
  (PR [#4249](https://github.com/pipecat-ai/pipecat/pull/4249))

- Added `MistralTTSService` for streaming text-to-speech using Mistral's
  Voxtral TTS API (`voxtral-mini-tts-2603`). Supports SSE-based audio streaming
  with automatic resampling from the API's native 24kHz to any requested sample
  rate. Requires the `mistral` optional extra (`pip install
  pipecat-ai[mistral]`).
  (PR [#4251](https://github.com/pipecat-ai/pipecat/pull/4251))

- Added `truncate_large_values` parameter to `LLMContext.get_messages()`. When
  `True`, returns compact deep copies of messages with binary data (base64
  images, audio) replaced by short placeholders and long string values in
  LLM-specific messages recursively truncated. Useful for serialization,
  logging, and debugging tools.
  (PR [#4272](https://github.com/pipecat-ai/pipecat/pull/4272))

- `CartesiaSTTService` now supports runtime settings updates (e.g. changing
  `language` or `model` via `STTUpdateSettingsFrame`). The service
  automatically reconnects with the new parameters. Previously, settings
  updates were silently ignored.
  (PR [#4282](https://github.com/pipecat-ai/pipecat/pull/4282))

- Added `pcm_32000` and `pcm_48000` sample rate support to ElevenLabs TTS
  services.
  (PR [#4293](https://github.com/pipecat-ai/pipecat/pull/4293))

- Added `enable_logging` parameter to `ElevenLabsHttpTTSService`. Set to
  `False` to enable zero retention mode (enterprise only).
  (PR [#4293](https://github.com/pipecat-ai/pipecat/pull/4293))

### Changed

- Updated `onnxruntime` from 1.23.2 to 1.24.3, adding support for Python 3.14.
  (PR [#3984](https://github.com/pipecat-ai/pipecat/pull/3984))

- MCPClient now requires async with MCPClient(...) as mcp: or explicit
  start()/close() calls to manage the connection lifecycle.
  (PR [#4034](https://github.com/pipecat-ai/pipecat/pull/4034))

- ⚠️ Updated `langchain` extra to require langchain 1.x (from 0.3.x),
  langchain-community 0.4.x (from 0.3.x), and langchain-openai 1.x (from
  0.3.x). If you pin these packages in your project, update your pins
  accordingly.
  (PR [#4192](https://github.com/pipecat-ai/pipecat/pull/4192))

- `WebsocketService` reconnection errors are now non-fatal. When a websocket
  service exhausts its reconnection attempts (either via exponential backoff or
  quick failure detection), it emits a non-fatal `ErrorFrame` instead of a
  fatal one. This allows application-level failover (e.g. `ServiceSwitcher`) to
  handle the failure instead of killing the entire pipeline.
  (PR [#4201](https://github.com/pipecat-ai/pipecat/pull/4201))

- Changed `GrokLLMService` default model from `grok-3-beta` to `grok-3`, now
  that the model is generally available.
  (PR [#4209](https://github.com/pipecat-ai/pipecat/pull/4209))

- `GoogleImageGenService` now defaults to `imagen-4.0-generate-001` (previously
  `imagen-3.0-generate-002`).
  (PR [#4213](https://github.com/pipecat-ai/pipecat/pull/4213))

- ⚠️ `BaseOpenAILLMService.get_chat_completions()` now accepts an `LLMContext`
  instead of `OpenAILLMInvocationParams`. If you override this method, update
  your signature accordingly.
  (PR [#4215](https://github.com/pipecat-ai/pipecat/pull/4215))

- When multiple function calls are returned in a single LLM response, by
  default (when `group_parallel_tools=True`) the LLM is now triggered exactly
  once after the last call in the batch completes, rather than waiting for all
  function calls.
  (PR [#4217](https://github.com/pipecat-ai/pipecat/pull/4217))

- ⚠️ `LLMService.function_call_timeout_secs` now defaults to `None` instead of
  `10.0`. Deferred function calls will run indefinitely unless a timeout is
  explicitly set at the service level or per-call. If you relied on the
  previous 10-second default, pass `function_call_timeout_secs=10.0`
  explicitly.
  (PR [#4224](https://github.com/pipecat-ai/pipecat/pull/4224))

- Updated `NvidiaTTSService`:

  - Made `api_key` optional for local NIM deployments.
  - Voice, language, and quality can be updated without reconnecting the gRPC
    client; new values take effect on the next synthesis turn, not for the
    current turn's in-flight requests.
  - Replaced per-sentence synchronous `synthesize_online` calls with async
    queue-backed gRPC streaming.
  - Streaming now uses asyncio tasks with explicit gRPC cancellation on
    interruption and stale-response filtering when a stream is aborted or
    replaced.
  - Renamed Riva references to Nemotron Speech in docs and messages.
  - Disabled automatic TTS start frames at the service level
    (`push_start_frame=False`) and emit `TTSStartedFrame` when a stitched
    synthesis stream is started for a context.
  (PR [#4249](https://github.com/pipecat-ai/pipecat/pull/4249))

### Removed

- ⚠️ Removed `OpenPipeLLMService` and the `openpipe` extra. OpenPipe was
  acquired by CoreWeave and the package is no longer maintained. If you were
  using `openpipe` as an LLM provider, switch to the underlying provider
  directly (e.g. `openai`). The OpenPipe interface can still be used with
  `OpenAILLMService` by specifying a `base_url`.
  (PR [#4191](https://github.com/pipecat-ai/pipecat/pull/4191))

- ⚠️ Removed `NoisereduceFilter`. Use system-level noise reduction or a
  service-based alternative instead.
  (PR [#4204](https://github.com/pipecat-ai/pipecat/pull/4204))

- ⚠️ Removed deprecated `vad_enabled` and `vad_audio_passthrough` transport
  params.
  (PR [#4204](https://github.com/pipecat-ai/pipecat/pull/4204))

- ⚠️ Removed deprecated `camera_in_enabled`, `camera_in_is_live`,
  `camera_in_width`, `camera_in_height`, `camera_out_enabled`,
  `camera_out_is_live`, `camera_out_width`, `camera_out_height`, and
  `camera_out_color` transport params. Use the `video_in_*` and `video_out_*`
  equivalents instead.
  (PR [#4204](https://github.com/pipecat-ai/pipecat/pull/4204))

- ⚠️ Removed `FrameProcessor.wait_for_task()`. Use `create_task()` and manage
  tasks with the built-in `TaskManager` instead.
  (PR [#4204](https://github.com/pipecat-ai/pipecat/pull/4204))

- ⚠️ Removed deprecated transport frames: `TransportMessageFrame`,
  `TransportMessageUrgentFrame`, `InputTransportMessageUrgentFrame`,
  `DailyTransportMessageFrame`, and `DailyTransportMessageUrgentFrame`. Use
  `OutputTransportMessageFrame`, `OutputTransportMessageUrgentFrame`,
  `InputTransportMessageFrame`, `DailyOutputTransportMessageFrame`, and
  `DailyOutputTransportMessageUrgentFrame` instead.
  (PR [#4204](https://github.com/pipecat-ai/pipecat/pull/4204))

- ⚠️ Removed `create_default_resampler()` from `pipecat.audio.utils`.
  (PR [#4204](https://github.com/pipecat-ai/pipecat/pull/4204))

- ⚠️ Removed `DailyRunner.configure_with_args()`. Use `PipelineRunner` with
  `RunnerArguments` instead.
  (PR [#4204](https://github.com/pipecat-ai/pipecat/pull/4204))

- ⚠️ Removed deprecated `on_pipeline_ended`, `on_pipeline_cancelled`, and
  `on_pipeline_stopped` events from `PipelineTask`. Use `on_pipeline_finished`
  instead.
  (PR [#4204](https://github.com/pipecat-ai/pipecat/pull/4204))

- ⚠️ Removed single-argument function call support from `LLMService`. Functions
  must use named parameters instead of a single `arguments` parameter.
  (PR [#4204](https://github.com/pipecat-ai/pipecat/pull/4204))

- ⚠️ Removed `FalSmartTurnAnalyzer` and `LocalSmartTurnAnalyzer`.
  (PR [#4204](https://github.com/pipecat-ai/pipecat/pull/4204))

- ⚠️ Removed `RTVIObserver.errors_enabled` parameter.
  (PR [#4204](https://github.com/pipecat-ai/pipecat/pull/4204))

- ⚠️ Removed deprecated RTVI models, frames, and processor methods including
  `RTVIConfig`, `RTVIServiceConfig`, `RTVIServiceOptionConfig`, various
  `RTVI*Data` models, `RTVIActionFrame`, and
  `RTVIProcessor.handle_function_call`/`handle_function_call_start`. Use the
  updated RTVI processor API instead.
  (PR [#4204](https://github.com/pipecat-ai/pipecat/pull/4204))

- ⚠️ Removed deprecated `KeypadEntryFrame` alias.
  (PR [#4204](https://github.com/pipecat-ai/pipecat/pull/4204))

- ⚠️ Removed deprecated interruption frames: `StartInterruptionFrame` and
  `BotInterruptionFrame`. Use `InterruptionFrame` and `InterruptionTaskFrame`
  instead.
  (PR [#4204](https://github.com/pipecat-ai/pipecat/pull/4204))

- ⚠️ Removed `LLMService.request_image_frame()`. Push a `UserImageRequestFrame`
  instead.
  (PR [#4204](https://github.com/pipecat-ai/pipecat/pull/4204))

- ⚠️ Removed `TTSService.say()`. Push a `TTSSpeakFrame` into the pipeline
  instead.
  (PR [#4204](https://github.com/pipecat-ai/pipecat/pull/4204))

- ⚠️ Removed `KrispFilter`. The `krisp` extra has been removed from
  `pyproject.toml`.
  (PR [#4204](https://github.com/pipecat-ai/pipecat/pull/4204))

- ⚠️ Removed `AudioBufferProcessor.user_continuous_stream` parameter. Use
  `user_audio_passthrough` instead.
  (PR [#4204](https://github.com/pipecat-ai/pipecat/pull/4204))

- ⚠️ Removed `LLMService.start_callback` parameter. Register an
  `on_llm_response_start` event handler instead.
  (PR [#4204](https://github.com/pipecat-ai/pipecat/pull/4204))

- ⚠️ Removed deprecated `observers` field from `PipelineParams`. Pass observers
  directly to `PipelineTask` constructor instead.
  (PR [#4204](https://github.com/pipecat-ai/pipecat/pull/4204))

- ⚠️ Removed deprecated `pipecat.services.openai_realtime` package. Use
  `pipecat.services.openai.realtime` instead.
  (PR [#4208](https://github.com/pipecat-ai/pipecat/pull/4208))

- ⚠️ Removed deprecated `pipecat.services.google.llm_vertex` module. Use
  `pipecat.services.google.vertex.llm` instead.
  (PR [#4208](https://github.com/pipecat-ai/pipecat/pull/4208))

- ⚠️ Removed deprecated `GoogleLLMOpenAIBetaService` from
  `pipecat.services.google.openai`. Use `GoogleLLMService` from
  `pipecat.services.google.llm` instead.
  (PR [#4208](https://github.com/pipecat-ai/pipecat/pull/4208))

- ⚠️ Removed deprecated `OpenAIRealtimeBetaLLMService` and
  `AzureRealtimeBetaLLMService`. Use `OpenAIRealtimeLLMService` and
  `AzureRealtimeLLMService` from `pipecat.services.openai.realtime` and
  `pipecat.services.azure.realtime` instead.
  (PR [#4208](https://github.com/pipecat-ai/pipecat/pull/4208))

- ⚠️ Removed deprecated `pipecat.services.ai_services` module. Import from
  `pipecat.services.ai_service`, `pipecat.services.llm_service`,
  `pipecat.services.stt_service`, `pipecat.services.tts_service`, etc. instead.
  (PR [#4208](https://github.com/pipecat-ai/pipecat/pull/4208))

- ⚠️ Removed deprecated `pipecat.services.gemini_multimodal_live` package. Use
  `pipecat.services.google.gemini_live` instead. Note that class names no
  longer include "Multimodal" (e.g. `GeminiMultimodalLiveLLMService` →
  `GeminiLiveLLMService`).
  (PR [#4208](https://github.com/pipecat-ai/pipecat/pull/4208))

- ⚠️ Removed deprecated `pipecat.services.google.gemini_live.llm_vertex`
  module. Use `pipecat.services.google.gemini_live.vertex.llm` instead.
  (PR [#4208](https://github.com/pipecat-ai/pipecat/pull/4208))

- ⚠️ Removed deprecated `pipecat.services.nim` package. Use
  `pipecat.services.nvidia.llm` instead (`NimLLMService` → `NvidiaLLMService`).
  (PR [#4208](https://github.com/pipecat-ai/pipecat/pull/4208))

- ⚠️ Removed deprecated `pipecat.services.deepgram.stt_sagemaker` and
  `pipecat.services.deepgram.tts_sagemaker` modules. Use
  `pipecat.services.deepgram.sagemaker.stt` and
  `pipecat.services.deepgram.sagemaker.tts` instead.
  (PR [#4208](https://github.com/pipecat-ai/pipecat/pull/4208))

- ⚠️ Removed deprecated `pipecat.services.aws_nova_sonic` package. Use
  `pipecat.services.aws.nova_sonic` instead.
  (PR [#4208](https://github.com/pipecat-ai/pipecat/pull/4208))

- ⚠️ Removed deprecated `pipecat.services.riva` package. Use
  `pipecat.services.nvidia.stt` and `pipecat.services.nvidia.tts` instead
  (`RivaSTTService` → `NvidiaSTTService`, `RivaTTSService` →
  `NvidiaTTSService`).
  (PR [#4208](https://github.com/pipecat-ai/pipecat/pull/4208))

- ⚠️ Removed deprecated compatibility modules:
  `pipecat.services.openai_realtime_beta` (use
  `pipecat.services.openai.realtime`),
  `pipecat.services.openai_realtime.context`,
  `pipecat.services.openai_realtime.frames`,
  `pipecat.services.openai.realtime.context`,
  `pipecat.services.openai.realtime.frames`,
  `pipecat.services.gemini_multimodal_live` (use
  `pipecat.services.google.gemini_live`),
  `pipecat.services.aws_nova_sonic.context` (use
  `pipecat.services.aws.nova_sonic`), `pipecat.services.google.openai` and
  `pipecat.services.google.llm_openai` (use `pipecat.services.google.llm`).
  (PR [#4215](https://github.com/pipecat-ai/pipecat/pull/4215))

- ⚠️ Removed `VisionImageFrameAggregator` (from
  `pipecat.processors.aggregators.vision_image_frame`). Vision/image handling
  is now built into `LLMContext` (from
  `pipecat.processors.aggregators.llm_context`). See the `12*` examples for the
  recommended replacement pattern.
  (PR [#4215](https://github.com/pipecat-ai/pipecat/pull/4215))

- ⚠️ Removed `OpenAILLMContext`, `OpenAILLMContextFrame`, and
  `OpenAILLMContext.from_messages()`. Use `LLMContext` (from
  `pipecat.processors.aggregators.llm_context`) and `LLMContextFrame` (from
  `pipecat.frames.frames`) instead. All services now exclusively use the
  universal `LLMContext`.

  From the developer's point of view, migrating will usually be a matter of
  going from this:

    ```python
    context = OpenAILLMContext(messages, tools)
    context_aggregator = llm.create_context_aggregator(context)
    ```

    To this:

    ```python
    from pipecat.processors.aggregators.llm_context import LLMContext
    from pipecat.processors.aggregators.llm_response_universal import LLMContextAggregatorPair

    context = LLMContext(messages, tools)
    context_aggregator = LLMContextAggregatorPair(context)
    ```
  (PR [#4215](https://github.com/pipecat-ai/pipecat/pull/4215))

- ⚠️ Removed deprecated frame types `LLMMessagesFrame` and
  `OpenAILLMContextAssistantTimestampFrame` from `pipecat.frames.frames`.
  Instead of `LLMMessagesFrame`, use `LLMContextFrame` with the new messages,
  or `LLMMessagesUpdateFrame` with `run_llm=True`.
  (PR [#4215](https://github.com/pipecat-ai/pipecat/pull/4215))

- ⚠️ Removed `GatedOpenAILLMContextAggregator` (from
  `pipecat.processors.aggregators.gated_open_ai_llm_context`). Use
  `GatedLLMContextAggregator` (from
  `pipecat.processors.aggregators.gated_llm_context`) instead.
  (PR [#4215](https://github.com/pipecat-ai/pipecat/pull/4215))

- ⚠️ Removed deprecated service-specific context and aggregator machinery,
  which was superseded by the universal `LLMContext` system.

  Service-specific classes removed: `AnthropicLLMContext`,
  `AnthropicContextAggregatorPair`, `AWSBedrockLLMContext`,
  `AWSBedrockContextAggregatorPair`, `OpenAIContextAggregatorPair`, and their
  user/assistant aggregators. Also removed `create_context_aggregator()` from
  `LLMService`, `OpenAILLMService`, `AnthropicLLMService`, and
  `AWSBedrockLLMService`.

  Base aggregator classes removed (from
  `pipecat.processors.aggregators.llm_response`): `BaseLLMResponseAggregator`,
  `LLMContextResponseAggregator`, `LLMUserContextAggregator`,
  `LLMAssistantContextAggregator`, `LLMUserResponseAggregator`,
  `LLMAssistantResponseAggregator`.

  From the developer's point of view, migrating will usually be a matter of
  going from this:

    ```python
    context = OpenAILLMContext(messages, tools)
    context_aggregator = llm.create_context_aggregator(context)
    ```

    To this:

    ```python
    from pipecat.processors.aggregators.llm_context import LLMContext
    from pipecat.processors.aggregators.llm_response_universal import LLMContextAggregatorPair

    context = LLMContext(messages, tools)
    context_aggregator = LLMContextAggregatorPair(context)
    ```
  (PR [#4215](https://github.com/pipecat-ai/pipecat/pull/4215))

- ⚠️ Removed deprecated service parameters and shims that have been replaced by
  the `settings=Service.Settings(...)` pattern or direct `__init__` parameters:
    - `PollyTTSService` alias (use `AWSTTSService`)
    - `TTSService`: `text_aggregator`, `text_filter` init params
    - `AWSNovaSonicLLMService`: `send_transcription_frames` init param
    - `DeepgramSTTService`: `url` init param (use `base_url`)
    - `FishAudioTTSService`: `model` init param (use `reference_id` or
      `settings`)
    - `GladiaSTTService`: `language` and `confidence` from `GladiaInputParams`,
      `InputParams` class alias
    - `GeminiTTSService`: `api_key` init param
    - `GeminiLiveLLMService`: `base_url` init param (use `http_options`)
    - `GoogleVertexLLMService`: `InputParams` class with
      `location`/`project_id` fields (use direct init params); `project_id` is now
      required, `location` defaults to `"us-east4"`
    - `MiniMaxHttpTTSService`: `english_normalization` from `InputParams` (use
      `text_normalization`)
    - `SimliVideoService`: `simli_config` init param (use `api_key`/`face_id`),
      `use_turn_server` init param; `api_key` and `face_id` are now required
    - `AnthropicLLMService`: `enable_prompt_caching_beta` from `InputParams`
      (use `enable_prompt_caching`)
  (PR [#4220](https://github.com/pipecat-ai/pipecat/pull/4220))

- ⚠️ Removed deprecated `pipecat.transports.services` and
  `pipecat.transports.network` module aliases. Update imports to use
  `pipecat.transports.daily.transport`, `pipecat.transports.livekit.transport`,
  `pipecat.transports.websocket.*`, `pipecat.transports.webrtc.*`, and
  `pipecat.transports.daily.utils` respectively.
  (PR [#4225](https://github.com/pipecat-ai/pipecat/pull/4225))

- ⚠️ Removed deprecated `pipecat.sync` package. Use `pipecat.utils.sync`
  instead.
  (PR [#4225](https://github.com/pipecat-ai/pipecat/pull/4225))

- ⚠️ Removed deprecated `TranscriptionMessage`, `ThoughtTranscriptionMessage`,
  and `TranscriptionUpdateFrame` from `pipecat.frames.frames`.
  (PR [#4228](https://github.com/pipecat-ai/pipecat/pull/4228))

- ⚠️ Removed deprecated `allow_interruptions` parameter from `PipelineParams`,
  `StartFrame`, and `FrameProcessor`. Interruptions are now always allowed by
  default. Use `LLMUserAggregator`'s `user_turn_strategies` /
  `user_mute_strategies` parameters to control interruption behavior.
  (PR [#4228](https://github.com/pipecat-ai/pipecat/pull/4228))

- ⚠️ Removed deprecated `STTMuteFilter`, `STTMuteConfig`, and `STTMuteStrategy`
  from `pipecat.processors.filters.stt_mute_filter`. Use
  `pipecat.turns.user_mute` strategies with `LLMUserAggregator`'s
  `user_mute_strategies` parameter instead.
  (PR [#4228](https://github.com/pipecat-ai/pipecat/pull/4228))

- ⚠️ Removed deprecated `pipecat.processors.transcript_processor` module
  (`TranscriptProcessor`, `TranscriptProcessorConfig`). Use pipeline observers
  instead.
  (PR [#4228](https://github.com/pipecat-ai/pipecat/pull/4228))

- ⚠️ Removed deprecated `EmulateUserStartedSpeakingFrame` and
  `EmulateUserStoppedSpeakingFrame` frames, and the `emulated` field from
  `UserStartedSpeakingFrame` / `UserStoppedSpeakingFrame`.
  (PR [#4228](https://github.com/pipecat-ai/pipecat/pull/4228))

- ⚠️ Removed deprecated `interruption_strategies` parameter from
  `PipelineParams`, `StartFrame`, and `FrameProcessor`. Use
  `LLMUserAggregator`'s `user_turn_strategies` parameter instead.
  (PR [#4228](https://github.com/pipecat-ai/pipecat/pull/4228))

- ⚠️ Removed deprecated `pipecat.audio.interruptions` module
  (`BaseInterruptionStrategy`, `MinWordsInterruptionStrategy`). Use
  `pipecat.turns.user_start.MinWordsUserTurnStartStrategy` with
  `LLMUserAggregator`'s `user_turn_strategies` parameter instead.
  (PR [#4228](https://github.com/pipecat-ai/pipecat/pull/4228))

- ⚠️ Removed deprecated `pipecat.utils.tracing.class_decorators` module. Use
  `pipecat.utils.tracing.service_decorators` instead.
  (PR [#4228](https://github.com/pipecat-ai/pipecat/pull/4228))

- ⚠️ Removed deprecated `add_pattern_pair` method from `PatternPairAggregator`.
  Use `add_pattern` instead.
  (PR [#4228](https://github.com/pipecat-ai/pipecat/pull/4228))

- ⚠️ Removed deprecated `UserResponseAggregator` class from
  `pipecat.processors.aggregators.user_response`. Use `LLMUserAggregator`
  instead.
  (PR [#4228](https://github.com/pipecat-ai/pipecat/pull/4228))

- ⚠️ Removed `ExternalUserTurnStrategies` and the automatic fallback to it in
  `LLMUserAggregator` when a `SpeechControlParamsFrame` was received from the
  transport.
  (PR [#4229](https://github.com/pipecat-ai/pipecat/pull/4229))

- ⚠️ Removed `vad_analyzer` and `turn_analyzer` parameters from
  `TransportParams` and all transport input classes, along with all deprecated
  VAD/turn analysis logic in `BaseInputTransport`. VAD and turn detection are
  now handled entirely by `LLMUserAggregator`.
  (PR [#4229](https://github.com/pipecat-ai/pipecat/pull/4229))

- ⚠️ Removed deprecated `TranscriptionUserTurnStopStrategy` alias (deprecated
  in 0.0.102). Use `SpeechTimeoutUserTurnStopStrategy` instead.
  (PR [#4232](https://github.com/pipecat-ai/pipecat/pull/4232))

- ⚠️ Removed deprecated `vad_events` setting and `should_interrupt` parameter
  from `DeepgramSTTService` (deprecated in 0.0.99). Use Silero VAD for voice
  activity detection instead.
  (PR [#4232](https://github.com/pipecat-ai/pipecat/pull/4232))

- ⚠️ Removed deprecated `send_transcription_frames` parameter from
  `OpenAIRealtimeLLMService` (deprecated in 0.0.92). Transcription frames are
  always sent.
  (PR [#4232](https://github.com/pipecat-ai/pipecat/pull/4232))

- ⚠️ Removed deprecated `UserIdleProcessor` (deprecated in 0.0.100). Use
  `LLMUserAggregator` with the `user_idle_timeout` parameter instead.
  (PR [#4232](https://github.com/pipecat-ai/pipecat/pull/4232))

- ⚠️ Removed deprecated `UserBotLatencyLogObserver` (deprecated in 0.0.102).
  Use `UserBotLatencyObserver` with its `on_latency_measured` event handler
  instead.
  (PR [#4232](https://github.com/pipecat-ai/pipecat/pull/4232))

- ⚠️ Removed the `riva` install extra. Use `nvidia` instead (`pip install
  "pipecat-ai[nvidia]"`).
  (PR [#4235](https://github.com/pipecat-ai/pipecat/pull/4235))

- Removed the empty `remote-smart-turn` install extra (was already a no-op).
  (PR [#4235](https://github.com/pipecat-ai/pipecat/pull/4235))

- ⚠️ Removed `DeprecatedModuleProxy` and all service `__init__.py` re-export
  shims. Flat imports like `from pipecat.services.openai import
  OpenAILLMService` no longer work. Use the full submodule path instead: `from
  pipecat.services.openai.llm import OpenAILLMService`. This is already the
  established pattern across all examples and internal code.
  (PR [#4239](https://github.com/pipecat-ai/pipecat/pull/4239))

- ⚠️ Removed deprecated `PIPECAT_OBSERVER_FILES` environment variable support.
  Use `PIPECAT_SETUP_FILES` instead.
  (PR [#4267](https://github.com/pipecat-ai/pipecat/pull/4267))

### Fixed

- Fixed `IdleFrameProcessor` where `asyncio.Event` was unconditionally cleared
  in a `finally` block instead of only on the success path.
  (PR [#3796](https://github.com/pipecat-ai/pipecat/pull/3796))

- Fixed MCPClient opening a new connection for every tool call instead of
  reusing the session.
  (PR [#4034](https://github.com/pipecat-ai/pipecat/pull/4034))

- GoogleLLMService now applies a low-latency thinking default
  (`thinking_level="minimal"`) for Gemini 3+ Flash models.
  (PR [#4067](https://github.com/pipecat-ai/pipecat/pull/4067))

- Fixed `WebsocketService` entering an infinite reconnection loop when a server
  accepts the WebSocket handshake but immediately closes the connection (e.g.
  invalid API key, close code 1008). The service now detects connections that
  fail repeatedly within seconds of being established and stops retrying after
  3 consecutive quick failures.
  (PR [#4201](https://github.com/pipecat-ai/pipecat/pull/4201))

- Fixed `InworldHttpTTSService` streaming responses crashing with
  `UnicodeDecodeError` when multi-byte UTF-8 characters were split across chunk
  boundaries. This caused TTS audio to cut off mid-sentence intermittently.
  (PR [#4202](https://github.com/pipecat-ai/pipecat/pull/4202))

- Fixed a crash (`JSONDecodeError`) when a user interruption occurs while the
  LLM is streaming function call arguments. Previously, the incomplete JSON
  arguments were passed directly to `json.loads()`, causing an unhandled
  exception. Affected services: OpenAI, Google (OpenAI-compatible), and
  SambaNova.
  (PR [#4203](https://github.com/pipecat-ai/pipecat/pull/4203))

- Fixed `BaseOutputTransport` discarding pending `UninterruptibleFrame` items
  (e.g. function-call context updates) when an interruption arrived. The audio
  task is now kept alive and only interruptible frames are drained when
  uninterruptible frames are present in the queue.
  (PR [#4217](https://github.com/pipecat-ai/pipecat/pull/4217))

- Fixed spurious LLM inference being triggered when a function call result
  arrived while the user was actively speaking. The context frame is now
  suppressed until the user stops speaking.
  (PR [#4217](https://github.com/pipecat-ai/pipecat/pull/4217))

- Fixed `CartesiaTTSService` failing with "Context has closed" errors when
  switching voice, model, or language via `TTSUpdateSettingsFrame`. The service
  now automatically flushes the current audio context and opens a fresh one
  when these settings change.
  (PR [#4220](https://github.com/pipecat-ai/pipecat/pull/4220))

- Fixed duplicate LLM replies that could occur when multiple async function
  call results arrived while an LLM request was already queued.
  (PR [#4230](https://github.com/pipecat-ai/pipecat/pull/4230))

- Fixed undefined `_warn_deprecated_param` calls in `OpenAIRealtimeLLMService`
  and `GrokRealtimeLLMService` for the deprecated `session_properties` init
  parameter.
  (PR [#4232](https://github.com/pipecat-ai/pipecat/pull/4232))

- Fixed Gemini Live bot hanging after a session resumption reconnect. Audio,
  video, and text input were silently dropped after reconnecting because the
  internal `_ready_for_realtime_input` flag was not being reset.
  (PR [#4242](https://github.com/pipecat-ai/pipecat/pull/4242))

- Fixed `VADController` getting stuck in the `SPEAKING` state when audio frames
  stop arriving mid-speech (e.g. user mutes mic). A new `audio_idle_timeout`
  parameter (default 1s, set to 0 to disable) forces a transition back to
  `QUIET` and emits `on_speech_stopped` when no audio is received while
  speaking.
  (PR [#4244](https://github.com/pipecat-ai/pipecat/pull/4244))

- Fixed `PipelineRunner._gc_collect()` blocking the event loop by running
  `gc.collect()` synchronously. Now offloaded via `asyncio.to_thread` to avoid
  stalling concurrent pipeline tasks.
  (PR [#4255](https://github.com/pipecat-ai/pipecat/pull/4255))

- Fixed `ElevenLabsTTSService` incorrectly enabling `auto_mode` when using
  `TextAggregationMode.TOKEN`. Auto mode disables server-side buffering and is
  designed for complete sentences — enabling it with token streaming degraded
  speech quality. The default is now derived automatically from the aggregation
  strategy: `auto_mode=True` for `SENTENCE`, `auto_mode=False` for `TOKEN`.
  Callers can still override by passing `auto_mode` explicitly.
  (PR [#4265](https://github.com/pipecat-ai/pipecat/pull/4265))

- Fixed `ValueError: write to closed file` during pipeline shutdown when
  observers were active. Observer proxy tasks are now cancelled before observer
  resources are cleaned up.
  (PR [#4267](https://github.com/pipecat-ai/pipecat/pull/4267))

- Fixed delayed turn completion when STT transcripts arrive after the p99
  timeout. Previously, a late transcript (beyond the p99 window) would fall
  through to the 5-second `user_turn_stop_timeout` fallback. Now the turn stop
  triggers immediately when the late transcript arrives.
  (PR [#4283](https://github.com/pipecat-ai/pipecat/pull/4283))

- Fixed `ElevenLabsTTSService` ignoring `enable_logging=False` and
  `enable_ssml_parsing=False`. The truthy check treated `False` the same as
  `None` (both skipped), and Python's `str(False)` produced `"False"` instead
  of the lowercase `"false"` expected by the API.
  (PR [#4293](https://github.com/pipecat-ai/pipecat/pull/4293))

- Fixed `on_assistant_turn_stopped` not resetting internal state when the LLM
  returned no text tokens. Added `interrupted` field to
  `AssistantTurnStoppedMessage` to indicate whether the assistant turn was
  interrupted.
  (PR [#4294](https://github.com/pipecat-ai/pipecat/pull/4294))

- Fixed `LLMContextSummarizer` failing with "No messages to summarize" when
  using `system_instruction` instead of a system-role message at the start of
  the context. The summarizer previously scanned the entire context for the
  first system message, which could match a mid-conversation injection (e.g.
  idle notifications) instead of the initial prompt, causing the summarization
  range to be empty.
  (PR [#4295](https://github.com/pipecat-ai/pipecat/pull/4295))

## [0.0.108] - 2026-03-27

### Added

- Added `SarvamLLMService` with support for `sarvam-30b`, `sarvam-30b-16k`,
  `sarvam-105b` and `sarvam-105b-32k`.
  (PR [#3978](https://github.com/pipecat-ai/pipecat/pull/3978))

- Added `on_turn_context_created(context_id)` hook to `TTSService`. Override
  this to perform provider-specific setup (e.g. eagerly opening a server-side
  context) before text starts flowing. Called each time a new turn context ID
  is created.
  (PR [#4013](https://github.com/pipecat-ai/pipecat/pull/4013))

- Added `XAIHttpTTSService` for text-to-speech using xAI's HTTP TTS API.
  (PR [#4031](https://github.com/pipecat-ai/pipecat/pull/4031))

- Added support for "developer" role messages in conversation context across
  all LLM adapters. For non-OpenAI services (Anthropic, Google, AWS Bedrock),
  "developer" messages are converted to "user" messages (use
  `system_instruction` to set the system instruction). For OpenAI services,
  "developer" messages pass through in conversation history. For the Responses
  API, they are kept as "developer" role (matching the existing "system" →
  "developer" conversion).
  (PR [#4089](https://github.com/pipecat-ai/pipecat/pull/4089))

- Added `SmallestTTSService`, a WebSocket-based TTS service integration with
  Smallest AI's Waves API. Supports the Lightning v2 and v3.1 models with
  configurable voice, language, speed, consistency, similarity, and enhancement
  settings.
  (PR [#4092](https://github.com/pipecat-ai/pipecat/pull/4092))

- Added warnings in turn stop strategies when `VADParams.stop_secs` differs
  from the recommended default (0.2s) or when `stop_secs >= STT p99 latency`,
  which collapses the STT wait timeout to 0s and may cause delayed turn
  detection. The warnings guide developers to re-run the
  [stt-benchmark](https://github.com/pipecat-ai/stt-benchmark) with their VAD
  settings.
  (PR [#4115](https://github.com/pipecat-ai/pipecat/pull/4115))

- Added `domain` parameter to `AssemblyAISTTSettings` for specialized
  recognition modes such as Medical Mode (`domain="medical-v1"`).
  (PR [#4117](https://github.com/pipecat-ai/pipecat/pull/4117))

- Added `NovitaLLMService` for using Novita AI's LLM models via their
  OpenAI-compatible API.
  (PR [#4119](https://github.com/pipecat-ai/pipecat/pull/4119))

- Added `cleanup()` method to `VADAnalyzer` and `VADController` so VAD analyzer
  resources are properly released when no longer needed. Custom `VADAnalyzer`
  subclasses can override `cleanup()` to free any held resources.
  (PR [#4120](https://github.com/pipecat-ai/pipecat/pull/4120))

- Added `on_end_of_turn` event handler to `AssemblyAISTTService`. This fires
  after the final transcript is pushed, providing a reliable hook for
  end-of-turn logic that doesn't race with `TranscriptionFrame`. Works in both
  Pipecat and AssemblyAI turn detection modes.
  (PR [#4128](https://github.com/pipecat-ai/pipecat/pull/4128))

- Added `DeepgramFluxSageMakerSTTService` for running Deepgram Flux
  speech-to-text on AWS SageMaker endpoints.  Use with
  `ExternalUserTurnStrategies` to take advantage of Flux's turn detection.
  (PR [#4143](https://github.com/pipecat-ai/pipecat/pull/4143))

- Added `Mem0MemoryService.get_memories()` convenience method for retrieving
  all stored memories outside the pipeline (e.g. to build a personalized
  greeting at connection time). This avoids the need to manually handle client
  type branching, filter construction, and async wrapping.
  (PR [#4156](https://github.com/pipecat-ai/pipecat/pull/4156))

### Changed

- Added context prewarming path for `InworldTTSService` to improve first audio
  latency.
  (PR [#4013](https://github.com/pipecat-ai/pipecat/pull/4013))

- Added `KrispVivaVadAnalyzer` for Voice Activity Detection using the Krisp
  VIVA SDK (requires `krisp_audio`).
  (PR [#4022](https://github.com/pipecat-ai/pipecat/pull/4022))

- Modified `InworldTTSService` to close context at end of turn instead of
  relying on idle timeout.
  (PR [#4028](https://github.com/pipecat-ai/pipecat/pull/4028))

- Added Gemini 3 support to the Gemini Live service.
  (PR [#4078](https://github.com/pipecat-ai/pipecat/pull/4078))

- `TTSService`: the default `stop_frame_timeout_s` (idle time before an
  automatic `TTSStoppedFrame` is pushed when `push_stop_frames=True`) has
  changed from `2.0` to `3.0` seconds.
  (PR [#4084](https://github.com/pipecat-ai/pipecat/pull/4084))

- ⚠️ `GeminiLLMAdapter` now only treats `messages[0]` as the initial system
  message, matching all other adapters. Previously it searched for the first
  "system" message anywhere in the conversation history. A "system" message
  appearing later in the list will now be converted to "user" instead of being
  extracted as the system instruction.
  (PR [#4089](https://github.com/pipecat-ai/pipecat/pull/4089))

- Fixed `InworldTtsService` to fallback to full text when TTS timestamps are
  not received.
  (PR [#4113](https://github.com/pipecat-ai/pipecat/pull/4113))

- ⚠️ Realtime services (Gemini Live, OpenAI Realtime, Grok Realtime, Nova
  Sonic) now prefer `system_instruction` from service settings over an initial
  system message in the LLM context, matching the behavior of non-realtime
  services. Previously, context-provided system instructions took precedence. A
  warning is now logged when both are set.
  (PR [#4130](https://github.com/pipecat-ai/pipecat/pull/4130))

- Bumped `nvidia-riva-client` minimum version to `>=2.25.1`.
  (PR [#4136](https://github.com/pipecat-ai/pipecat/pull/4136))

- Upgraded `protobuf` from 5.x to 6.x (`>=6.31.1,<7`).
  (PR [#4136](https://github.com/pipecat-ai/pipecat/pull/4136))

- Unrecognized language strings (e.g. Deepgram's `"multi"`) no longer produce a
  warning at startup. The log message has been downgraded to debug level since
  these are valid service-specific values that are passed through correctly.
  (PR [#4137](https://github.com/pipecat-ai/pipecat/pull/4137))

- `GrokLLMService` and `GrokRealtimeLLMService` now live in the
  `pipecat.services.xai` module alongside `XAIHttpTTSService`, since all three
  use the same xAI API. Update imports from `pipecat.services.grok.*` to
  `pipecat.services.xai.*` (e.g. `from pipecat.services.xai.llm import
  GrokLLMService`).
  (PR [#4142](https://github.com/pipecat-ai/pipecat/pull/4142))

- ⚠️ Bumped `mem0ai` dependency from `~=0.1.94` to `>=1.0.8,<2`. Users of the
  `mem0` extra will need to update their mem0ai package.
  (PR [#4156](https://github.com/pipecat-ai/pipecat/pull/4156))

### Deprecated

- `pipecat.services.grok.llm`, `pipecat.services.grok.realtime.llm`, and
  `pipecat.services.grok.realtime.events` are deprecated. The old import paths
  still work but emit a `DeprecationWarning`; use `pipecat.services.xai.llm`,
  `pipecat.services.xai.realtime.llm`, and
  `pipecat.services.xai.realtime.events` instead.
  (PR [#4142](https://github.com/pipecat-ai/pipecat/pull/4142))

### Removed

- ⚠️ `TTSService.add_word_timestamps()` no longer supports the `"Reset"` and
  `"TTSStoppedFrame"` sentinel strings. If you have a custom TTS service that
  called `await self.add_word_timestamps([("Reset", 0)])` or `await
  self.add_word_timestamps([("TTSStoppedFrame", 0), ("Reset", 0)], ctx_id)`,
  replace them with `await self.append_to_audio_context(ctx_id,
  TTSStoppedFrame(context_id=ctx_id))` and let `_handle_audio_context` manage
  the word-timestamp reset automatically.
  (PR [#4145](https://github.com/pipecat-ai/pipecat/pull/4145))

- Removed `SambaNovaSTTService`. SambaNova no longer offers speech-to-text
  audio models. Use another STT provider instead.
  (PR [#4154](https://github.com/pipecat-ai/pipecat/pull/4154))

### Fixed

- Fixed Gemini Live (`GoogleGeminiLiveLLMService`) not honoring
  `settings.system_instruction`. The system instruction was being read from a
  deprecated constructor parameter instead of the settings object, causing it
  to be silently ignored.
  (PR [#4089](https://github.com/pipecat-ai/pipecat/pull/4089))

- Fixed `AWSBedrockLLMAdapter` sending an empty message list to the API when
  the only message in context was a system message. The lone system message is
  now converted to "user" role instead of being extracted, matching the
  existing Anthropic adapter behavior.
  (PR [#4089](https://github.com/pipecat-ai/pipecat/pull/4089))

- Fixed Gemini Live pipeline hanging indefinitely when an `EndFrame` was
  deferred while waiting for the bot to finish responding and `turn_complete`
  never arrived. As a possible root-cause fix, `turn_complete` messages are now
  handled even if they lack `usage_metadata`. As a fallback, the deferred
  `EndFrame` now has a 30-second safety timeout.
  (PR [#4125](https://github.com/pipecat-ai/pipecat/pull/4125))

- Fixed ElevenLabs WebSocket disconnections (1008 "Maximum simultaneous
  contexts exceeded") caused by rapid user interruptions. When interruptions
  arrived before any TTS text was generated, phantom contexts were created on
  the ElevenLabs server that were never closed, eventually exceeding the
  5-context limit.
  (PR [#4126](https://github.com/pipecat-ai/pipecat/pull/4126))

- Fixed the final sentence being dropped from the conversation context when
  using RTVI text input with non-word-timestamp TTS services. The
  `LLMFullResponseEndFrame` was racing ahead of the last `TTSTextFrame`,
  causing the `LLMAssistantAggregator` to finalize the context before the final
  sentence arrived.
  (PR [#4127](https://github.com/pipecat-ai/pipecat/pull/4127))

- Fixed audio crackling and popping in recordings when both user and bot are
  speaking. `AudioBufferProcessor` no longer injects silence into a track's
  buffer while that track is actively producing audio, preventing mid-utterance
  interruptions in the recorded output.
  (PR [#4135](https://github.com/pipecat-ai/pipecat/pull/4135))

- Fixed websocket TTS word timestamps so interrupted contexts cannot leak stale
  words or backward PTS values into later turns.
  (PR [#4145](https://github.com/pipecat-ai/pipecat/pull/4145))

- Fixed a race condition in `InterruptibleTTSService` where, if `run_tts` had
  been invoked but `BotStartedSpeakingFrame` had not yet been received, a user
  interruption could allow stale audio to leak through.
  (PR [#4145](https://github.com/pipecat-ai/pipecat/pull/4145))

- Fixed Gemini Live local VAD mode (`GeminiVADParams(disabled=True)` with
  external VAD) not working. The bot now correctly detects user speech and
  signals turn boundaries to the Gemini API.
  (PR [#4146](https://github.com/pipecat-ai/pipecat/pull/4146))

- Fixed Gemini Live message handling to process all `server_content` fields
  independently. Gemini 3.x can bundle multiple fields (e.g. `model_turn` and
  `output_transcription`) on the same message, but the previous `elif` chain
  only processed the first match, silently dropping the rest.
  (PR [#4147](https://github.com/pipecat-ai/pipecat/pull/4147))

- Fixed `ServiceSwitcher` with `ServiceSwitcherStrategyFailover` incorrectly
  triggering failover when `ErrorFrame`s from other pipeline stages (e.g. TTS)
  propagated upstream through the switcher. Previously, any non-fatal error
  passing through would be misattributed to the active service and trigger an
  unwanted service switch. Now only errors originating from the switcher's own
  managed services trigger failover.
  (PR [#4149](https://github.com/pipecat-ai/pipecat/pull/4149))

- Fixed `LiveKitOutputTransport` not clearing the `rtc.AudioSource` internal
  buffer on interruption, causing the bot to continue speaking for several
  seconds after being interrupted.
  (PR [#4151](https://github.com/pipecat-ai/pipecat/pull/4151))

- Fixed a crash in OpenAI LLM processing when the provider returns
  `chunk.choices[0].delta.audio = None`, which caused `'NoneType' object has no
  attribute 'get'` errors during audio transcript handling.
  (PR [#4152](https://github.com/pipecat-ai/pipecat/pull/4152))

- Fixed error floods in `DeepgramSTTService` when the WebSocket connection
  drops. With Deepgram SDK 6.x, `send_media()` raises exceptions on a dead
  connection instead of silently failing, causing every queued audio frame to
  log an error. Now `send_media()` failures are caught gracefully — a single
  warning is logged and audio frames are skipped until the existing
  reconnection logic restores the connection.
  (PR [#4153](https://github.com/pipecat-ai/pipecat/pull/4153))

- `Mem0MemoryService` no longer blocks the event loop during memory storage and
  retrieval. All Mem0 API calls now run in a background thread, and message
  storage is fire-and-forget so it doesn't delay downstream processing.
  (PR [#4156](https://github.com/pipecat-ai/pipecat/pull/4156))

- Fixed `Mem0MemoryService` failing to store messages when the context
  contained system or developer role messages. The Mem0 API only accepts user
  and assistant roles, so other roles are now filtered out before storing.
  (PR [#4156](https://github.com/pipecat-ai/pipecat/pull/4156))

- Added missing `on_dtmf_event` callback to `LemonSliceTransportClient.setup()`
  `DailyCallbacks` construction, fixing a `ValidationError` at pipeline setup
  time.
  (PR [#4161](https://github.com/pipecat-ai/pipecat/pull/4161))

- Fixed an issue in `InworldTTSService` where, in cases of fast interruption,
  we would continue receiving audio from the previous context.
  (PR [#4167](https://github.com/pipecat-ai/pipecat/pull/4167))

- Fixed a word timestamp interleaving issue in `InworldTTSService` when
  processing multiple sentences.
  (PR [#4167](https://github.com/pipecat-ai/pipecat/pull/4167))

- Fixed duplicate `TTSStoppedFrame` being pushed in TTS services using
  `push_stop_frames=True`. When the stop-frame timeout fired, a second
  `TTSStoppedFrame` could be pushed after the normal one at context completion.
  (PR [#4172](https://github.com/pipecat-ai/pipecat/pull/4172))

- ⚠️ Fixed `DeepgramSTTService` compatibility with deepgram-sdk 6.1.0. The SDK
  now requires explicit message objects for `send_keep_alive()`,
  `send_close_stream()`, and `send_finalize()`. The minimum deepgram-sdk
  version is now 6.1.0.
  (PR [#4174](https://github.com/pipecat-ai/pipecat/pull/4174))

- Fixed RTVI events not being delivered to clients when using WebSocket
  transports. `ProtobufFrameSerializer` now sets `ignore_rtvi_messages=False`
  by default.
  (PR [#4176](https://github.com/pipecat-ai/pipecat/pull/4176))

- Fixed a timing issue where turn detection timer tasks (idle controller,
  speech timeout, turn analyzer, and turn completion) could miss their first
  tick because the newly created asyncio task was not yet scheduled when the
  caller continued.
  (PR [#4183](https://github.com/pipecat-ai/pipecat/pull/4183))

- Fixed `FastAPIWebsocketTransport` intermittently hanging on shutdown when the
  remote side (e.g. Twilio) disconnects while audio is being sent. A race
  condition between the send and receive paths could cause the
  `on_client_disconnected` callback to be skipped, leaving the pipeline waiting
  for a disconnect signal that never came.
  (PR [#4186](https://github.com/pipecat-ai/pipecat/pull/4186))

### Performance

- `RimeTTSService` now handles Rime's `done` WebSocket message to complete
  audio contexts immediately, eliminating the 3-second idle timeout that
  previously added latency at the end of each utterance.
  (PR [#4172](https://github.com/pipecat-ai/pipecat/pull/4172))

## [0.0.107] - 2026-03-23

### Added

- Added `frame_order` parameter to `SyncParallelPipeline`. Set
  `frame_order=FrameOrder.PIPELINE` to push synchronized output frames in
  pipeline definition order (all frames from the first pipeline, then the
  second, etc.) instead of the default arrival order.
  (PR [#4029](https://github.com/pipecat-ai/pipecat/pull/4029))

- Added `sync_with_audio` field to `OutputImageRawFrame`. When set to `True`,
  the output transport queues image frames with audio so they are displayed
  only after all preceding audio has been sent, enabling synchronized
  audio/image playback.
  (PR [#4029](https://github.com/pipecat-ai/pipecat/pull/4029))

- Added `OpenAIResponsesLLMService`, a new LLM service that uses the OpenAI
  Responses API. Supports streaming text, function calling, usage metrics, and
  out-of-band inference. Works with the universal `LLMContext` and
  `LLMContextAggregatorPair`. See
  `examples/foundational/07-interruptible-openai-responses.py` and
  `14-function-calling-openai-responses.py`.
  (PR [#4074](https://github.com/pipecat-ai/pipecat/pull/4074))

- Added `audio_out_auto_silence` parameter to `TransportParams` (defaults to
  `True`). When set to `False`, the transport waits for audio data instead of
  inserting silence when the output queue is empty, which is useful for
  scenarios that require uninterrupted audio playback without artificial gaps.
  (PR [#4104](https://github.com/pipecat-ai/pipecat/pull/4104))

### Changed

- Renamed tracing span attributes to align with OpenTelemetry GenAI semantic
  conventions: `gen_ai.system` to `gen_ai.provider.name`, `system` to
  `gen_ai.system_instructions`, `gen_ai.usage.cache_read_input_tokens` to
  `gen_ai.usage.cache_read.input_tokens`, and
  `gen_ai.usage.cache_creation_input_tokens` to
  `gen_ai.usage.cache_creation.input_tokens`.
  (PR [#3449](https://github.com/pipecat-ai/pipecat/pull/3449))

- `DeepgramSageMakerTTSService` now correctly routes audio through the base
  `TTSService` audio context queue. Audio frames are delivered via
  `append_to_audio_context()` instead of being pushed directly, enabling proper
  ordering, interruption handling, and start/stop frame lifecycle management.
  Interruptions now trigger a `Clear` message to Deepgram (flushing its text
  buffer) at the right time via `on_audio_context_interrupted`.
  (PR [#4083](https://github.com/pipecat-ai/pipecat/pull/4083))

- `GradiumTTSService` now sends a per-context `setup` message with
  `client_req_id` before the first text message for each TTS context, following
  Gradium's multiplexing protocol. Previously, a single setup message was sent
  at connection time without a `client_req_id`, which prevented Gradium from
  associating requests with their sessions when using `close_ws_on_eos=False`.
  (PR [#4091](https://github.com/pipecat-ai/pipecat/pull/4091))

### Fixed

- Fixed stale `system_instruction` in LLM tracing spans by reading from
  `_settings.system_instruction` instead of the removed `_system_instruction`
  attribute.
  (PR [#3449](https://github.com/pipecat-ai/pipecat/pull/3449))

- Fixed `SyncParallelPipeline` breaking the Whisker debugger.
  (PR [#4029](https://github.com/pipecat-ai/pipecat/pull/4029))

- Fixed `SyncParallelPipeline` race condition where concurrent SystemFrame
  processing (e.g. from RTVI) could corrupt sink queues and cause deadlocks.
  SystemFrames now take a fast path that passes them through without draining
  queued output.
  (PR [#4029](https://github.com/pipecat-ai/pipecat/pull/4029))

- Fixed TTS frame ordering so that non-system frames always arrive in correct
  order relative to the `TTSStartedFrame`/`TTSAudioRawFrame`/`TTSStoppedFrame`
  sequence. Previously these frames could race ahead of or behind audio context
  frames, producing out-of-order output downstream.
  (PR [#4075](https://github.com/pipecat-ai/pipecat/pull/4075))

- Fixed `SarvamTTSService` audio and error frames now route through
  `append_to_audio_context()` instead of `push_frame()`, ensuring correct
  behavior with audio contexts and interruptions.
  (PR [#4082](https://github.com/pipecat-ai/pipecat/pull/4082))

- Fixed audio frame ordering and interruption handling in Fish Audio, LMNT,
  Neuphonic, and Rime NonJson TTS services. These services were bypassing the
  base `TTSService` audio context serialization queue by pushing audio frames
  directly, which could cause out-of-order frames and broken interruptions
  during speech.
  (PR [#4090](https://github.com/pipecat-ai/pipecat/pull/4090))

- Fixed Genesys AudioHook serializer to always include the `parameters` field in
  protocol messages. The AudioHook protocol requires every message to carry a
  `parameters` object (even if empty), but `_create_message` omitted it when no
  parameters were provided. This caused clients that validate message structure
  (including the Genesys reference implementation) to reject `pong` and
  parameter-less `closed` responses, breaking server sequence tracking and
  preventing `outputVariables` from reaching the Architect flow.
  (PR [#4093](https://github.com/pipecat-ai/pipecat/pull/4093))

## [0.0.106] - 2026-03-18

### Added

- Added optional `service` field to `ServiceUpdateSettingsFrame` (and its
  subclasses `LLMUpdateSettingsFrame`, `TTSUpdateSettingsFrame`,
  `STTUpdateSettingsFrame`) to target a specific service instance. When
  `service` is set, only the matching service applies the settings; others
  forward the frame unchanged. This enables updating a single service when
  multiple services of the same type exist in the pipeline.
  (PR [#4004](https://github.com/pipecat-ai/pipecat/pull/4004))

- Added `sip_provider` and `room_geo` parameters to `configure()` in the Daily
  runner. These convenience parameters let callers specify a SIP provider name
  and geographic region directly without manually constructing
  `DailyRoomProperties` and `DailyRoomSipParams`.
  (PR [#4005](https://github.com/pipecat-ai/pipecat/pull/4005))

- Added `PerplexityLLMAdapter` that automatically transforms conversation
  messages to satisfy Perplexity's stricter API constraints (strict role
  alternation, no non-initial system messages, last message must be user/tool).
  Previously, certain conversation histories could cause Perplexity API errors
  that didn't occur with OpenAI (`PerplexityLLMService` subclasses
  `OpenAILLMService` since Perplexity uses an OpenAI-compatible API).
  (PR [#4009](https://github.com/pipecat-ai/pipecat/pull/4009))

- Added DTMF input event support to the Daily transport. Incoming DTMF tones
  are now received via Daily's `on_dtmf_event` callback and pushed into the
  pipeline as `InputDTMFFrame`, enabling bots to react to keypad presses from
  phone callers.
  (PR [#4047](https://github.com/pipecat-ai/pipecat/pull/4047))

- Added `WakePhraseUserTurnStartStrategy` for triggering user turns based on
  wake phrases, with support for `single_activation` mode. Deprecates
  `WakeCheckFilter`.
  (PR [#4064](https://github.com/pipecat-ai/pipecat/pull/4064))

- Added `default_user_turn_start_strategies()` and
  `default_user_turn_stop_strategies()` helper functions for composing custom
  strategy lists.
  (PR [#4064](https://github.com/pipecat-ai/pipecat/pull/4064))

### Changed

- Changed tool result JSON serialization to use `ensure_ascii=False`,
  preserving UTF-8 characters instead of escaping them. This reduces context
  size and token usage for non-English languages.
  (PR [#3457](https://github.com/pipecat-ai/pipecat/pull/3457))

- `OpenAIRealtimeSTTService`'s `noise_reduction` parameter is now part of
  `OpenAIRealtimeSTTSettings`, making it runtime-updatable via
  `STTUpdateSettingsFrame`. The direct `noise_reduction` init argument is
  deprecated as of 0.0.106.
  (PR [#3991](https://github.com/pipecat-ai/pipecat/pull/3991))

- Updated `sarvamai` dependency from `0.1.26a2` (alpha) to `0.1.26` (stable
  release).
  (PR [#3997](https://github.com/pipecat-ai/pipecat/pull/3997))

- `SimliVideoService` now extends `AIService` instead of `FrameProcessor`,
  aligning it with the HeyGen and Tavus video services. It supports
  `SimliVideoService.Settings(...)` for configuration and uses
  `start()`/`stop()`/`cancel()` lifecycle methods. Existing constructor usage
  (`api_key`, `face_id`, etc.) remains unchanged.
  (PR [#4001](https://github.com/pipecat-ai/pipecat/pull/4001))

- Update `pipecat-ai-small-webrtc-prebuilt` to `2.4.0`.
  (PR [#4023](https://github.com/pipecat-ai/pipecat/pull/4023))

- Nova Sonic assistant text transcripts are now delivered in real-time using
  speculative text events instead of delayed final text events. Previously,
  assistant text only arrived after all audio had finished playing, causing
  laggy transcripts in client UIs. Speculative text arrives before each audio
  chunk, providing text synchronized with what the bot is saying. This also
  simplifies the internal text handling by removing the interruption re-push
  hack and assistant text buffer.
  (PR [#4042](https://github.com/pipecat-ai/pipecat/pull/4042))

- Updated `daily-python` dependency to 0.25.0.
  (PR [#4047](https://github.com/pipecat-ai/pipecat/pull/4047))

- Added `enable_dialout` parameter to `configure()` in `pipecat.runner.daily`
  to support dial-out rooms. Also narrowed misleading `Optional` type hints and
  deduplicated token expiry calculation.
  (PR [#4048](https://github.com/pipecat-ai/pipecat/pull/4048))

- Extended `ProcessFrameResult` to stop strategies, allowing a stop strategy to
  short-circuit evaluation of subsequent strategies by returning `STOP`.
  (PR [#4064](https://github.com/pipecat-ai/pipecat/pull/4064))

- `GradiumSTTService` now takes both an `encoding` and `sample_rate`
  constructor argument which is assmebled in the class to form the
  `input_format`. PCM accepts `8000`, `16000`, and `24000` Hz sample rates.
  (PR [#4066](https://github.com/pipecat-ai/pipecat/pull/4066))

- Improved `GradiumSTTService` transcription accuracy by reworking how text
  fragments are accumulated and finalized. Previously, trailing words could be
  dropped when the server's `flushed` response arrived before all text tokens
  were delivered. The service now uses a short aggregation delay after flush to
  capture trailing tokens, producing complete utterances.
  (PR [#4066](https://github.com/pipecat-ai/pipecat/pull/4066))

### Deprecated

- `SimliVideoService.InputParams` is deprecated. Use the direct constructor
  parameters `max_session_length`, `max_idle_time`, and `enable_logging`
  instead.
  (PR [#4001](https://github.com/pipecat-ai/pipecat/pull/4001))

- Deprecated `LocalSmartTurnAnalyzerV2` and `LocalCoreMLSmartTurnAnalyzer`. Use
  `LocalSmartTurnAnalyzerV3` instead. Instantiating these analyzers will now
  emit a `DeprecationWarning`.
  (PR [#4012](https://github.com/pipecat-ai/pipecat/pull/4012))

- Deprecated `WakeCheckFilter` in favor of `WakePhraseUserTurnStartStrategy`.
  (PR [#4064](https://github.com/pipecat-ai/pipecat/pull/4064))

### Fixed

- Fixed an issue where the default model for `OpenAILLMService` and
  `AzureLLMService` was mistakenly reverted to `gpt-4o`. The defaults are now
  restored to `gpt-4.1`.
  (PR [#4000](https://github.com/pipecat-ai/pipecat/pull/4000))

- Fixed a race condition where `EndTaskFrame` could cause the pipeline to shut
  down before in-flight frames (e.g. LLM function call responses) finished
  processing. `EndTaskFrame` and `StopTaskFrame` now flow through the pipeline
  as `ControlFrame`s, ensuring all pending work is flushed before shutdown
  begins. `CancelTaskFrame` and `InterruptionTaskFrame` remain immediate
  (`SystemFrame`).
  (PR [#4006](https://github.com/pipecat-ai/pipecat/pull/4006))

- Fixed `ParallelPipeline` dropping or misordering frames during lifecycle
  synchronization. Buffered frames are now flushed in the correct order
  relative to synchronization frames (`StartFrame` goes first,
  `EndFrame`/`CancelFrame` go after), and frames added to the buffer during
  flush are also drained.
  (PR [#4007](https://github.com/pipecat-ai/pipecat/pull/4007))

- Fixed `TTSService` potentially canceling in-flight audio during shutdown. The
  stop sequence now waits for all queued audio contexts to finish processing
  before canceling the stop frame task.
  (PR [#4007](https://github.com/pipecat-ai/pipecat/pull/4007))

- Fixed `Language` enum values (e.g. `Language.ES`) not being converted to
  service-specific codes when passed via
  `settings=Service.Settings(language=Language.ES)` at init time. This caused
  API errors (e.g. 400 from Rime) because the raw enum was sent instead of the
  expected language code (e.g. `"spa"`). Runtime updates via
  `UpdateSettingsFrame` were unaffected. The fix centralizes conversion in the
  base `TTSService` and `STTService` classes so all services handle this
  consistently.
  (PR [#4024](https://github.com/pipecat-ai/pipecat/pull/4024))

- Fixed `DeepgramSTTService` ignoring the `base_url` scheme when using `ws://`
  or `http://`. Previously these were silently overwritten with `wss://` /
  `https://`, breaking air-gapped or private deployments that don't use TLS.
  All scheme choices (`wss://`, `https://`, `ws://`, `http://`, or bare
  hostname) are now respected.
  (PR [#4026](https://github.com/pipecat-ai/pipecat/pull/4026))

- Fixed `LLMSwitcher.register_function()` and `register_direct_function()` not
  accepting or forwarding the `timeout_secs` parameter.
  (PR [#4037](https://github.com/pipecat-ai/pipecat/pull/4037))

- Fixed empty user transcriptions in Nova Sonic causing spurious interruptions.
  Previously, an empty transcription could trigger an interruption of the
  assistant's response even though the user hadn't actually spoken.
  (PR [#4042](https://github.com/pipecat-ai/pipecat/pull/4042))

- Fixed `SonioxSTTService` and `OpenAIRealtimeSTTService` crash when language
  parameters contain plain strings instead of `Language` enum values.
  (PR [#4046](https://github.com/pipecat-ai/pipecat/pull/4046))

- Fixed premature user turn stops caused by late transcriptions arriving
  between turns. A stale transcript from the previous turn could persist into
  the next turn and trigger a stop before the current turn's real transcript
  arrived. Stop strategies are now reset at both turn start and turn stop to
  prevent state from leaking across turn boundaries.
  (PR [#4057](https://github.com/pipecat-ai/pipecat/pull/4057))

- Fixed raw language strings like `"de-DE"` silently failing when passed to
  TTS/STT services (e.g. ElevenLabs producing no audio). Raw strings now go
  through the same `Language` enum resolution as enum values, so regional codes
  like `"de-DE"` are properly converted to service-expected formats like
  `"de"`. Unrecognized strings log a warning instead of failing silently.
  (PR [#4058](https://github.com/pipecat-ai/pipecat/pull/4058))

- Fixed Deepgram STT list-type settings (`keyterm`, `keywords`, `search`,
  `redact`, `replace`) being stringified instead of passed as lists to the SDK,
  which caused them to be sent as literal strings (e.g. `"['pipecat']"`) in the
  WebSocket query params.
  (PR [#4063](https://github.com/pipecat-ai/pipecat/pull/4063))

- Fixed `MinWordsUserTurnStartStrategy` including text below the word threshold
  in the output by resetting aggregation when the minimum word count is not
  met.
  (PR [#4064](https://github.com/pipecat-ai/pipecat/pull/4064))

- Fixed audio overlap and potential dropped TTS content when multiple assistant
  turns occur in quick succession. `TTSService` now flushes remaining text
  before pausing frame processing on `LLMFullResponseEndFrame`/`EndFrame`,
  instead of pausing first.
  (PR [#4071](https://github.com/pipecat-ai/pipecat/pull/4071))

### Security

- Bumped PyJWT minimum version from 2.10.1 to 2.12.0 in the `livekit` extra to
  address CVE-2026-32597 (GHSA-752w-5fwx-jx9f), where PyJWT <= 2.11.0 accepted
  unknown `crit` header extensions.
  (PR [#4035](https://github.com/pipecat-ai/pipecat/pull/4035))

## [0.0.105] - 2026-03-10

### Added

- Added concurrent audio context support: `CartesiaTTSService` can now
  synthesize the next sentence while the previous one is still playing, by
  setting `pause_frame_processing=False` and routing each sentence through its
  own audio context queue.
  (PR [#3804](https://github.com/pipecat-ai/pipecat/pull/3804))

- Added custom video track support to Daily transport. Use
  `video_out_destinations` in `DailyParams` to publish multiple video tracks
  simultaneously, mirroring the existing `audio_out_destinations` feature.
  (PR [#3831](https://github.com/pipecat-ai/pipecat/pull/3831))

- Added `ServiceSwitcherStrategyFailover` that automatically switches to the
  next service when the active service reports a non-fatal error. Recovery
  policies can be implemented via the `on_service_switched` event handler.
  (PR [#3861](https://github.com/pipecat-ai/pipecat/pull/3861))

- Added optional `timeout_secs` parameter to `register_function()` and
  `register_direct_function()` for per-tool function call timeout control,
  overriding the global `function_call_timeout_secs` default.
  (PR [#3915](https://github.com/pipecat-ai/pipecat/pull/3915))

- Added `cloud-audio-only` recording option to Daily transport's
  `enable_recording` property.
  (PR [#3916](https://github.com/pipecat-ai/pipecat/pull/3916))

- Wired up `system_instruction` in `BaseOpenAILLMService`,
  `AnthropicLLMService`, and `AWSBedrockLLMService` so it works as a default
  system prompt, matching the behavior of the Google services. This enables
  sharing a single `LLMContext` across multiple LLM services, where each
  service provides its own system instruction independently.

    ```python
    llm = OpenAILLMService(
        api_key=os.getenv("OPENAI_API_KEY"),
        system_instruction="You are a helpful assistant.",
    )

    context = LLMContext()


    @transport.event_handler("on_client_connected")
    async def on_client_connected(transport, client):
        context.add_message({"role": "user", "content": "Please introduce yourself."})
        await task.queue_frames([LLMRunFrame()])
    ```
  (PR [#3918](https://github.com/pipecat-ai/pipecat/pull/3918))

- Added `vad_threshold` parameter to `AssemblyAIConnectionParams` for
  configuring voice activity detection sensitivity in U3 Pro. Aligning this
  with external VAD thresholds (e.g., Silero VAD) prevents the "dead zone"
  where AssemblyAI transcribes speech that VAD hasn't detected yet.
  (PR [#3927](https://github.com/pipecat-ai/pipecat/pull/3927))

- Added `push_empty_transcripts` parameter to `BaseWhisperSTTService` and
  `OpenAISTTService` to allow empty transcripts to be pushed downstream as
  `TranscriptionFrame` instead of discarding them (the default behavior). This
  is intended for situations where VAD fires even though the user did not
  speak. In these cases, it is useful to know that nothing was transcribed so
  that the agent can resume speaking, instead of waiting longer for a
  transcription.
  (PR [#3930](https://github.com/pipecat-ai/pipecat/pull/3930))

- LLM services (`BaseOpenAILLMService`, `AnthropicLLMService`,
  `AWSBedrockLLMService`) now log a warning when both `system_instruction` and
  a system message in the context are set. The constructor's
  `system_instruction` takes precedence.
  (PR [#3932](https://github.com/pipecat-ai/pipecat/pull/3932))

- Runtime settings updates (via `STTUpdateSettingsFrame`) now work for AWS
  Transcribe, Azure, Cartesia, Deepgram, ElevenLabs Realtime, Gradium, and
  Soniox STT services. Previously, changing settings at runtime only stored the
  new values without reconnecting.
  (PR [#3946](https://github.com/pipecat-ai/pipecat/pull/3946))

- Exposed `on_summary_applied` event on `LLMAssistantAggregator`, allowing
  users to listen for context summarization events without accessing private
  members.
  (PR [#3947](https://github.com/pipecat-ai/pipecat/pull/3947))

- Deepgram Flux STT settings (`keyterm`, `eot_threshold`,
  `eager_eot_threshold`, `eot_timeout_ms`) can now be updated mid-stream via
  `STTUpdateSettingsFrame` without triggering a reconnect. The new values are
  sent to Deepgram as a Configure WebSocket message on the existing connection.
  (PR [#3953](https://github.com/pipecat-ai/pipecat/pull/3953))

- Added `system_instruction` parameter to `run_inference` across all LLM
  services, allowing callers to override the system prompt for one-shot
  inference calls. Used by `_generate_summary` to pass the summarization prompt
  cleanly.
  (PR [#3968](https://github.com/pipecat-ai/pipecat/pull/3968))

### Changed

- Audio context management (previously in `AudioContextTTSService`) is now
  built into `TTSService`. All WebSocket providers (`cartesia`, `elevenlabs`,
  `asyncai`, `inworld`, `rime`, `gradium`, `resembleai`) now inherit from
  `WebsocketTTSService` directly. Word-timestamp baseline is set automatically
  on the first audio chunk of each context instead of requiring each provider
  to call `start_word_timestamps()` in their receive loop.
  (PR [#3804](https://github.com/pipecat-ai/pipecat/pull/3804))

- Daily transport now uses `CustomVideoSource`/`CustomVideoTrack` instead of
  `VirtualCameraDevice` for the default camera output, mirroring how audio
  already works with `CustomAudioSource`/`CustomAudioTrack`.
  (PR [#3831](https://github.com/pipecat-ai/pipecat/pull/3831))

- ⚠️ Updated `DeepgramSTTService` to use `deepgram-sdk` v6. The `LiveOptions`
  class was removed from the SDK and is now provided by pipecat directly;
  import it from `pipecat.services.deepgram.stt` instead of `deepgram`.
  (PR [#3848](https://github.com/pipecat-ai/pipecat/pull/3848))

- `ServiceSwitcherStrategy` base class now provides a `handle_error()` hook for
  subclasses to implement error-based switching. `ServiceSwitcher` defaults to
  `ServiceSwitcherStrategyManual` and `strategy_type` is now optional.
  (PR [#3861](https://github.com/pipecat-ai/pipecat/pull/3861))

- Support for Voice Focus 2.0 models.
    - Updated `aic-sdk` to `~=2.1.0` to support Voice Focus 2.0 models.
    - Cleaned unused `ParameterFixedError` exception handling in `AICFilter`
      parameter setup.
  (PR [#3889](https://github.com/pipecat-ai/pipecat/pull/3889))

- `max_context_tokens` and `max_unsummarized_messages` in
  `LLMAutoContextSummarizationConfig` (and deprecated
  `LLMContextSummarizationConfig`) can now be set to `None` independently to
  disable that summarization threshold. At least one must remain set.
  (PR [#3914](https://github.com/pipecat-ai/pipecat/pull/3914))

- ⚠️ Removed `formatted_finals` and `word_finalization_max_wait_time` from
  `AssemblyAIConnectionParams` as these were v2 API parameters not supported in
  v3. Clarified that `format_turns` only applies to Universal-Streaming models;
  U3 Pro has automatic formatting built-in.
  (PR [#3927](https://github.com/pipecat-ai/pipecat/pull/3927))

- Changed `DeepgramTTSService` to send a Clear message on interruption instead
  of disconnecting and reconnecting the WebSocket, allowing the connection to
  persist throughout the session.
  (PR [#3958](https://github.com/pipecat-ai/pipecat/pull/3958))

- Re-added `enhancement_level` support to `AICFilter` with runtime
  `FilterEnableFrame` control, applying `ProcessorParameter.Bypass` and
  `ProcessorParameter.EnhancementLevel` together.
  (PR [#3961](https://github.com/pipecat-ai/pipecat/pull/3961))

- Updated `daily-python` dependency from `~=0.23.0` to `~=0.24.0`.
  (PR [#3970](https://github.com/pipecat-ai/pipecat/pull/3970))

- Updated `FishAudioTTSService` default model from `s1` to `s2-pro`, matching
  Fish Audio's latest recommended model for improved quality and speed.
  (PR [#3973](https://github.com/pipecat-ai/pipecat/pull/3973))

- `AzureSTTService` `region` parameter is now optional when `private_endpoint`
  is provided. A `ValueError` is raised if neither is given, and a warning is
  logged if both are provided (`private_endpoint` takes priority).
  (PR [#3974](https://github.com/pipecat-ai/pipecat/pull/3974))

### Deprecated

- Deprecated `AudioContextTTSService` and `AudioContextWordTTSService`.
  Subclass `WebsocketTTSService` directly instead; audio context management is
  now part of the base `TTSService`.
  - Deprecated `WordTTSService`, `WebsocketWordTTSService`, and
    `InterruptibleWordTTSService`. Word timestamp logic is now always active in
    `TTSService` and no longer needs to be opted into via a subclass.
  (PR [#3804](https://github.com/pipecat-ai/pipecat/pull/3804))

- Deprecated `pipecat.services.google.llm_vertex`,
  `pipecat.services.google.llm_openai`, and
  `pipecat.services.google.gemini_live.llm_vertex` modules. Use
  `pipecat.services.google.vertex.llm`, `pipecat.services.google.openai.llm`,
  and `pipecat.services.google.gemini_live.vertex.llm` instead. The old import
  paths still work but will emit a `DeprecationWarning`.
  (PR [#3980](https://github.com/pipecat-ai/pipecat/pull/3980))

### Removed

- ⚠️ Removed `supports_word_timestamps` parameter from `TTSService.__init__()`.
  Word timestamp logic is now always active. Remove this argument from any
  custom subclass `super().__init__()` calls.
  (PR [#3804](https://github.com/pipecat-ai/pipecat/pull/3804))

### Fixed

- Fixed `DeepgramSTTService` keepalive ping timeout disconnections. The
  deepgram-sdk v6 removed automatic keepalive; pipecat now sends explicit
  `KeepAlive` messages every 5 seconds, within the recommended 3–5 second
  interval before Deepgram's 10-second inactivity timeout.
  (PR [#3848](https://github.com/pipecat-ai/pipecat/pull/3848))

- Fixed `BufferError: Existing exports of data: object cannot be re-sized` in
  `AICFilter` caused by holding a `memoryview` on the mutable audio buffer
  across async yield points.
  (PR [#3889](https://github.com/pipecat-ai/pipecat/pull/3889))

- Fixed TTS context not being appended to the assistant message history when
  using `TTSSpeakFrame` with `append_to_context=True` with some TTS providers.
  (PR [#3936](https://github.com/pipecat-ai/pipecat/pull/3936))

- Fixed context summarization leaving orphaned tool responses in the kept
  context when tool calls were moved to the summarized portion.
  (PR [#3937](https://github.com/pipecat-ai/pipecat/pull/3937))

- Fixed turn completion state not resetting at end of LLM responses.
  `LLMFullResponseEndFrame` is pushed (not received) by the LLM service, so the
  mixin now handles it in `push_frame` instead of `process_frame`.
  (PR [#3956](https://github.com/pipecat-ai/pipecat/pull/3956))

- Fixed turn completion instructions being injected as a context system message
  instead of using `system_instruction`. This caused warning spam when
  `system_instruction` was also set and didn't persist across full context
  updates.
  (PR [#3957](https://github.com/pipecat-ai/pipecat/pull/3957))

- Fixed `TTSService` audio context queue getting blocked when
  `append_to_audio_context()` was called with a `None` context ID, which
  prevented subsequent audio from being delivered.
  (PR [#3958](https://github.com/pipecat-ai/pipecat/pull/3958))

- Fixed `on_call_state_updated` event handler in LiveKit transport receiving
  incorrect number of arguments due to redundant `self` passed to
  `_call_event_handler`.
  (PR [#3959](https://github.com/pipecat-ai/pipecat/pull/3959))

- Fixed OpenAI Realtime, OpenAI Realtime Beta, and Grok realtime services
  treating `conversation_already_has_active_response` as a fatal error. These
  services now log it as a non-fatal debug event when a response is already in
  progress.
  (PR [#3960](https://github.com/pipecat-ai/pipecat/pull/3960))

- Fixed `SmallWebRTCConnection` silently discarding messages sent before the
  data channel is open by queuing them and flushing once the channel is ready.
  A bounded queue (`MAX_MESSAGE_QUEUE_SIZE = 50`) prevents unbounded memory
  growth, and a 10-second timeout after connection clears the queue and falls
  back to discard mode if the data channel never opens.
  (PR [#3962](https://github.com/pipecat-ai/pipecat/pull/3962))

- Fixed `AzureSTTService` failing to initialize when `private_endpoint` is
  provided. The Azure Speech SDK's `SpeechConfig` does not accept both `region`
  and `endpoint` simultaneously, so they are now passed conditionally.
  (PR [#3967](https://github.com/pipecat-ai/pipecat/pull/3967))

- Fixed `GoogleLLMService` ignoring the `system_instruction` set via
  constructor or `GoogleLLMSettings` when a system message was also present in
  the context. The settings value now correctly takes priority, and a warning
  is logged when both are set.
  (PR [#3976](https://github.com/pipecat-ai/pipecat/pull/3976))

### Other

- Updated foundational examples to use `system_instruction` on LLM services
  instead of adding system messages to `LLMContext`.
  (PR [#3918](https://github.com/pipecat-ai/pipecat/pull/3918))

- Updated AssemblyAI turn detection example to use `keyterms_prompt` list
  format instead of `prompt` string for improved clarity.
  (PR [#3929](https://github.com/pipecat-ai/pipecat/pull/3929))

- Updated foundational examples and eval scripts to use `"user"` role instead
  of `"system"` when adding messages to `LLMContext`, since system prompts
  should be set via `system_instruction` on the LLM service.
  (PR [#3931](https://github.com/pipecat-ai/pipecat/pull/3931))

## [0.0.104] - 2026-03-02

### Added

- Added `TextAggregationMetricsData` metric measuring the time from the first
  LLM token to the first complete sentence, representing the latency cost of
  sentence aggregation in the TTS pipeline.
  (PR [#3696](https://github.com/pipecat-ai/pipecat/pull/3696))

- Added support for using strongly-typed objects instead of dicts for updating
  service settings at runtime.

    Instead of, say:

    ```python
    await task.queue_frame(STTUpdateSettingsFrame(settings={"language": Language.ES}))
    ```

    you'd do:

    ```python
    await task.queue_frame(STTUpdateSettingsFrame(delta=DeepgramSTTSettings(language=Language.ES)))
    ```

  Each service now vends strongly-typed classes like `DeepgramSTTSettings`
  representing the service's runtime-updatable settings.
  (PR [#3714](https://github.com/pipecat-ai/pipecat/pull/3714))

- Added support for specifying private endpoints for Azure Speech-to-Text,
  enabling use in private networks behind firewalls.
  (PR [#3764](https://github.com/pipecat-ai/pipecat/pull/3764))

- Added `LemonSliceTransport` and `LemonSliceApi` to support adding real-time
  LemonSlice Avatars to any Daily room.
  (PR [#3791](https://github.com/pipecat-ai/pipecat/pull/3791))

- Added `output_medium` parameter to `AgentInputParams` and
  `OneShotInputParams` in Ultravox service to control initial output medium
  (text or voice) at call creation time.
  (PR [#3806](https://github.com/pipecat-ai/pipecat/pull/3806))

- Added `TurnMetricsData` as a generic metrics class for turn detection, with
  e2e processing time measurement. `KrispVivaTurn` now emits `TurnMetricsData`
  with `e2e_processing_time_ms` tracking the interval from VAD
  speech-to-silence transition to turn completion.
  (PR [#3809](https://github.com/pipecat-ai/pipecat/pull/3809))

- Added `on_audio_context_interrupted()` and `on_audio_context_completed()`
  callbacks to `AudioContextTTSService`. Subclasses can override these to
  perform provider-specific cleanup instead of overriding
  `_handle_interruption()`.
  (PR [#3814](https://github.com/pipecat-ai/pipecat/pull/3814))

- Added `on_summary_applied` event to `LLMContextSummarizer` for observability,
  providing message counts before and after context summarization.
  (PR [#3855](https://github.com/pipecat-ai/pipecat/pull/3855))

- Added `summary_message_template` to `LLMContextSummarizationConfig` for
  customizing how summaries are formatted when injected into context (e.g.,
  wrapping in XML tags).
  (PR [#3855](https://github.com/pipecat-ai/pipecat/pull/3855))

- Added `summarization_timeout` to `LLMContextSummarizationConfig` (default
  120s) to prevent hung LLM calls from permanently blocking future
  summarizations.
  (PR [#3855](https://github.com/pipecat-ai/pipecat/pull/3855))

- Added optional `llm` field to `LLMContextSummarizationConfig` for routing
  summarization to a dedicated LLM service (e.g., a cheaper/faster model)
  instead of the pipeline's primary model.
  (PR [#3855](https://github.com/pipecat-ai/pipecat/pull/3855))

- Add AssemblyAI u3-rt-pro model support with built-in turn detection mode
  (PR [#3856](https://github.com/pipecat-ai/pipecat/pull/3856))

- Added `LLMSummarizeContextFrame` to trigger on-demand context summarization
  from anywhere in the pipeline (e.g. a function call tool). Accepts an
  optional `config: LLMContextSummaryConfig` to override summary generation
  settings per request.
  (PR [#3863](https://github.com/pipecat-ai/pipecat/pull/3863))

- Added `LLMContextSummaryConfig` (summary generation params:
  `target_context_tokens`, `min_messages_after_summary`,
  `summarization_prompt`) and `LLMAutoContextSummarizationConfig` (auto-trigger
  thresholds: `max_context_tokens`, `max_unsummarized_messages`, plus a nested
  `summary_config`). These replace the monolithic
  `LLMContextSummarizationConfig`.
  (PR [#3863](https://github.com/pipecat-ai/pipecat/pull/3863))

- Added support for the `speed_alpha` parameter to the `arcana` model in
  `RimeTTSService`.
  (PR [#3873](https://github.com/pipecat-ai/pipecat/pull/3873))

- Added `ClientConnectedFrame`, a new `SystemFrame` pushed by all transports
  (Daily, LiveKit, FastAPI WebSocket, WebSocket Server, SmallWebRTC, HeyGen,
  Tavus) when a client connects. Enables observers to track transport readiness
  timing.
  (PR [#3881](https://github.com/pipecat-ai/pipecat/pull/3881))

- Added `StartupTimingObserver` for measuring how long each processor's
  `start()` method takes during pipeline startup. Also measures transport
  readiness — the time from `StartFrame` to first client connection — via the
  `on_transport_timing_report` event.
  (PR [#3881](https://github.com/pipecat-ai/pipecat/pull/3881))

- Added `BotConnectedFrame` for SFU transports and `on_transport_timing_report`
  event to `StartupTimingObserver` with bot and client connection timing.
  (PR [#3881](https://github.com/pipecat-ai/pipecat/pull/3881))

- Added optional `direction` parameter to `PipelineTask.queue_frame()` and
  `PipelineTask.queue_frames()`, allowing frames to be pushed upstream from the
  end of the pipeline.
  (PR [#3883](https://github.com/pipecat-ai/pipecat/pull/3883))

- Added `on_latency_breakdown` event to `UserBotLatencyObserver` providing
  per-service TTFB, text aggregation, user turn duration, and function call
  latency metrics for each user-to-bot response cycle.
  (PR [#3885](https://github.com/pipecat-ai/pipecat/pull/3885))

- Added `on_first_bot_speech_latency` event to `UserBotLatencyObserver`
  measuring the time from client connection to first bot speech. An
  `on_latency_breakdown` is also emitted for this first speech event.
  (PR [#3885](https://github.com/pipecat-ai/pipecat/pull/3885))

- Added `broadcast_interruption()` to `FrameProcessor`. This method pushes an
  `InterruptionFrame` both upstream and downstream directly from the calling
  processor, avoiding the round-trip through the pipeline task that
  `push_interruption_task_frame_and_wait()` required.
  (PR [#3896](https://github.com/pipecat-ai/pipecat/pull/3896))

### Changed

- Added `text_aggregation_mode` parameter to `TTSService` and all TTS
  subclasses with a new `TextAggregationMode` enum (`SENTENCE`, `TOKEN`). All
  text now flows through text aggregators regardless of mode, enabling pattern
  detection and tag handling in TOKEN mode.
  (PR [#3696](https://github.com/pipecat-ai/pipecat/pull/3696))

- ⚠️ Refactored runtime-updatable service settings to use strongly-typed
  classes (`TTSSettings`, `STTSettings`, `LLMSettings`, and service-specific
  subclasses) instead of plain dicts. Each service's `_settings` now holds
  these strongly-typed objects. For service maintainers, see changes in
  COMMUNITY_INTEGRATIONS.md.
  (PR [#3714](https://github.com/pipecat-ai/pipecat/pull/3714))

- Word timestamp support has been moved from `WordTTSService` into `TTSService`
  via a new `supports_word_timestamps` parameter. Services that previously
  extended `WordTTSService`, `AudioContextWordTTSService`, or
  `WebsocketWordTTSService` now pass `supports_word_timestamps=True` to their
  parent `__init__` instead.
  (PR [#3786](https://github.com/pipecat-ai/pipecat/pull/3786))

- Improved Ultravox TTFB measurement accuracy by using VAD speech end time
  instead of `UserStoppedSpeakingFrame` timing.
  (PR [#3806](https://github.com/pipecat-ai/pipecat/pull/3806))

- Aligned `UltravoxRealtimeLLMService` frame handling with OpenAI/Gemini
  realtime services: added `InterruptionFrame` handling with metrics cleanup,
  processing metrics at response boundaries, and improved agent transcript
  handling for both voice and text output modalities.
  (PR [#3806](https://github.com/pipecat-ai/pipecat/pull/3806))

- Updated `OpenAIRealtimeLLMService` default model to `gpt-realtime-1.5`.
  (PR [#3807](https://github.com/pipecat-ai/pipecat/pull/3807))

- Added `api_key` parameter to `KrispVivaSDKManager`, `KrispVivaTurn`, and
  `KrispVivaFilter` for Krisp SDK v1.6.1+ licensing. Falls back to
  `KRISP_VIVA_API_KEY` environment variable.
  (PR [#3809](https://github.com/pipecat-ai/pipecat/pull/3809))

- Bumped `nltk` minimum version from 3.9.1 to 3.9.3 to resolve a security
  vulnerability.
  (PR [#3811](https://github.com/pipecat-ai/pipecat/pull/3811))

- `ServiceSettingsUpdateFrame`s are now `UninterruptibleFrame`s. Generally
  speaking, you don't want a user interruption to prevent a service setting
  change from going into effect. Note that you usually don't use
  `ServiceSettingsUpdateFrame` directly, you use one of its subclasses:
    - `LLMUpdateSettingsFrame`
    - `TTSUpdateSettingsFrame`
    - `STTUpdateSettingsFrame`
  (PR [#3819](https://github.com/pipecat-ai/pipecat/pull/3819))

- Updated context summarization to use `user` role instead of `assistant` for
  summary messages.
  (PR [#3855](https://github.com/pipecat-ai/pipecat/pull/3855))

- Rename `AssemblyAISTTService` parameter
  `min_end_of_turn_silence_when_confident` parameter to `min_turn_silence` (old
  name still supported with deprecation warning)
  (PR [#3856](https://github.com/pipecat-ai/pipecat/pull/3856))

- ⚠️ Renamed `LLMAssistantAggregatorParams` fields:
  `enable_context_summarization` → `enable_auto_context_summarization` and
  `context_summarization_config` → `auto_context_summarization_config` (now
  accepts `LLMAutoContextSummarizationConfig`). The old names still work with a
  `DeprecationWarning` for one release cycle.
  (PR [#3863](https://github.com/pipecat-ai/pipecat/pull/3863))

- `ElevenLabsRealtimeSTTService` now sets `TranscriptionFrame.finalized` to
  `True` when using `CommitStrategy.MANUAL`.
  (PR [#3865](https://github.com/pipecat-ai/pipecat/pull/3865))

- Updated numba version pin from == to >=0.61.2
  (PR [#3868](https://github.com/pipecat-ai/pipecat/pull/3868))

- Updated tracing code to use `ServiceSettings` dataclass API
  (`given_fields()`, attribute access) instead of dict-style access
  (`.items()`, `in`, subscript).
  (PR [#3879](https://github.com/pipecat-ai/pipecat/pull/3879))

- ⚠️ Removed `event` field and `complete()` method from `InterruptionFrame`.
  Removed `event` field from `InterruptionTaskFrame`. These are no longer
  needed since `broadcast_interruption()` does not require a round-trip
  completion signal.
  (PR [#3896](https://github.com/pipecat-ai/pipecat/pull/3896))

- Moved `pipecat.services.deepgram.stt_sagemaker` and
  `pipecat.services.deepgram.tts_sagemaker` to
  `pipecat.services.deepgram.sagemaker.stt` and
  `pipecat.services.deepgram.sagemaker.tts`. The old import paths still work
  but emit a `DeprecationWarning`.
  (PR [#3902](https://github.com/pipecat-ai/pipecat/pull/3902))

### Deprecated

- ⚠️ Deprecated `aggregate_sentences` parameter on `TTSService` and all TTS
  subclasses. Use `text_aggregation_mode=TextAggregationMode.SENTENCE` or
  `text_aggregation_mode=TextAggregationMode.TOKEN` instead.
  (PR [#3696](https://github.com/pipecat-ai/pipecat/pull/3696))

- Deprecated `set_model()`, `set_voice()`, and `set_language()` on AI services
  in favor of runtime updates via `TTSUpdateSettingsFrame`,
  `STTUpdateSettingsFrame`, and `LLMUpdateSettingsFrame`.

  ⚠️ Note, too, a subtle behavior change in these deprecated methods. Whereas
  previously only `set_language()` caused the service to actually react to the
  update (e.g. by reconnecting to a remote service so it an pick up the
  change), now all these methods do. This change was made as part of a refactor
  making them all work the same way under the hood.
  (PR [#3714](https://github.com/pipecat-ai/pipecat/pull/3714))

- Dict-based `*UpdateSettingsFrame(settings={...})` is deprecated in favor of
  passing typed settings delta objects with
  `*UpdateSettingsFrame(delta={...})`.
  (PR [#3714](https://github.com/pipecat-ai/pipecat/pull/3714))

- Deprecated `WordTTSService`, `WebsocketWordTTSService`,
  `AudioContextWordTTSService`, and `InterruptibleWordTTSService`. Use their
  non-word counterparts with `supports_word_timestamps=True` instead:
    - `WordTTSService` → `TTSService(supports_word_timestamps=True)`
    - `WebsocketWordTTSService` →
  `WebsocketTTSService(supports_word_timestamps=True)`
    - `AudioContextWordTTSService` →
  `AudioContextTTSService(supports_word_timestamps=True)`
    - `InterruptibleWordTTSService` →
  `InterruptibleTTSService(supports_word_timestamps=True)`
  (PR [#3786](https://github.com/pipecat-ai/pipecat/pull/3786))

- Deprecated `SmartTurnMetricsData` in favor of `TurnMetricsData`.
  `BaseSmartTurn` now emits `TurnMetricsData` directly.
  (PR [#3809](https://github.com/pipecat-ai/pipecat/pull/3809))

- Deprecated `LLMContextSummarizationConfig`. Use
  `LLMAutoContextSummarizationConfig` with a nested `LLMContextSummaryConfig`
  instead. The old class emits a `DeprecationWarning`.
  (PR [#3863](https://github.com/pipecat-ai/pipecat/pull/3863))

- Deprecated `push_interruption_task_frame_and_wait()` in `FrameProcessor`. Use
  `broadcast_interruption()` instead. The old method now delegates to
  `broadcast_interruption()` and logs a deprecation warning.
  (PR [#3896](https://github.com/pipecat-ai/pipecat/pull/3896))

### Removed

- Removed `local-smart-turn-v3` optional extra from `pyproject.toml`. The
  `transformers` and `onnxruntime` packages are now always installed as core
  dependencies since they are required by the default turn stop strategy,
  `TurnAnalyzerUserTurnStopStrategy` which uses `LocalSmartTurnAnalyzerV3`.
  (PR [#3803](https://github.com/pipecat-ai/pipecat/pull/3803))

- ⚠️ Removed `PlayHTTTSService` and `PlayHTHttpTTSService`. PlayHT has been
  shut down and is no longer available.
  (PR [#3838](https://github.com/pipecat-ai/pipecat/pull/3838))

### Fixed

- Added `LLMSpecificMessage` handling in `LLMContextSummarizationUtil` to skip
  provider-specific messages during context summarization.
  (PR [#3794](https://github.com/pipecat-ai/pipecat/pull/3794))

- Treated `response_cancel_not_active` as a non-fatal error in realtime
  services (`OpenAIRealtimeLLMService`, `GrokRealtimeLLMService`,
  `OpenAIRealtimeBetaLLMService`) to prevent WebSocket disconnection when
  cancelling an inactive response.
  (PR [#3795](https://github.com/pipecat-ai/pipecat/pull/3795))

- Fixed Poetry compatibility by inlining `local-smart-turn-v3` dependencies
  (`transformers`, `onnxruntime`) into core dependencies instead of using a
  self-referential extra.
  (PR [#3803](https://github.com/pipecat-ai/pipecat/pull/3803))

- Fixed `SentryMetrics` method signatures to match updated
  `FrameProcessorMetrics` base class, resolving `TypeError` when using
  `start_time`/`end_time` keyword arguments.
  (PR [#3808](https://github.com/pipecat-ai/pipecat/pull/3808))

- Fixed STT TTFB metrics not being reported for `SonioxSTTService` and
  `AWSTranscribeSTTService` due to missing `can_generate_metrics()` override.
  (PR [#3813](https://github.com/pipecat-ai/pipecat/pull/3813))

- Fixed an issue where `AudioContextTTSService`-based providers (AsyncAI,
  ElevenLabs, Inworld, Rime) did not close or clean up their server-side audio
  contexts after normal speech completion, only on interruption.
  (PR [#3814](https://github.com/pipecat-ai/pipecat/pull/3814))

- Fixed STT TTFB metrics measuring timeout expiry time instead of actual
  transcript arrival time.
  (PR [#3822](https://github.com/pipecat-ai/pipecat/pull/3822))

- Fixed `InterimTranscriptionFrame` and `TranslationFrame` being
  unintentionally pushed downstream in `LLMUserAggregator`. They are now
  consumed like `TranscriptionFrame`.
  (PR [#3825](https://github.com/pipecat-ai/pipecat/pull/3825))

- Fixed misleading "Empty audio frame received for STT service" warnings when
  using audio filters (e.g. `RNNoiseFilter`, `KrispVivaFilter`, `AICFilter`)
  that buffer audio internally.
  (PR [#3828](https://github.com/pipecat-ai/pipecat/pull/3828))

- Fixed issues with `RimeNonJsonTTSService` where trailing punctuation is
  sometimes vocalized
  (PR [#3837](https://github.com/pipecat-ai/pipecat/pull/3837))

- Fixed `TTSSpeakFrame` not committing spoken text to the conversation context
  when used outside of an LLM response (e.g., bot greetings or injected
  speech).
  (PR [#3845](https://github.com/pipecat-ai/pipecat/pull/3845))

- Removed verbose per-chunk audio logging from `GenesysAudioHookSerializer`
  that flooded production logs.
  (PR [#3850](https://github.com/pipecat-ai/pipecat/pull/3850))

- Add beta feature warning when using custom prompts with AssemblyAI
  (PR [#3856](https://github.com/pipecat-ai/pipecat/pull/3856))

- Fixed `LocalSmartTurnAnalyzerV3` producing incorrect end-of-turn predictions
  at non-16kHz sample rates (e.g. 8kHz Twilio telephony) by adding automatic
  resampling to 16kHz before Whisper feature extraction.
  (PR [#3857](https://github.com/pipecat-ai/pipecat/pull/3857))

- Fixed `PipelineTask` double-inserting `RTVIProcessor` into the frame chain
  when the user provides both an `RTVIProcessor` in the pipeline and a custom
  `RTVIObserver` subclass in observers.
  (PR [#3867](https://github.com/pipecat-ai/pipecat/pull/3867))

- Fixed turn completion instructions being lost when `LLMMessagesUpdateFrame`
  replaces the LLM context. When `filter_incomplete_user_turns` is enabled, the
  turn completion system message is now re-injected after context replacement.
  (PR [#3888](https://github.com/pipecat-ai/pipecat/pull/3888))

- Fixed Azure TTS and STT services silently swallowing cancellation errors
  (invalid API key, network failures, rate limiting) instead of propagating
  them as `ErrorFrame`s to the pipeline.
  (PR [#3893](https://github.com/pipecat-ai/pipecat/pull/3893))

### Performance

- Switched `GradiumTTSService` from `InterruptibleWordTTSService` to
  `AudioContextWordTTSService`, eliminating websocket disconnect/reconnect on
  every interruption by using `client_req_id`-based multiplexing.
  (PR [#3759](https://github.com/pipecat-ai/pipecat/pull/3759))

### Other

- Standardized Sarvam STT/TTS User-Agent header handling to consistently send
  Pipecat SDK identity in websocket requests.
  (PR [#3886](https://github.com/pipecat-ai/pipecat/pull/3886))

## [0.0.103] - 2026-02-20

### Added

- Added `"timestampTransportStrategy": "ASYNC"` to `InworldAITTSService`. This
  allows timestamps info to trail audio chunks arrival, resulting in much
  better first audio chunk latency
  (PR [#3625](https://github.com/pipecat-ai/pipecat/pull/3625))

- Added model-specific `InputParams` to `RimeTTSService`: arcana params
  (`repetition_penalty`, `temperature`, `top_p`) and mistv2 params
  (`no_text_normalization`, `save_oovs`, `segment`). Model, voice, and param
  changes now trigger WebSocket reconnection.
  (PR [#3642](https://github.com/pipecat-ai/pipecat/pull/3642))

- Added `write_transport_frame()` hook to `BaseOutputTransport` allowing
  transport subclasses to handle custom frame types that flow through the audio
  queue.
  (PR [#3719](https://github.com/pipecat-ai/pipecat/pull/3719))

- Added `DailySIPTransferFrame` and `DailySIPReferFrame` to the Daily
  transport.  These frames queue SIP transfer and SIP REFER operations with
  audio, so the operation executes only after the bot finishes its current
  utterance.
  (PR [#3719](https://github.com/pipecat-ai/pipecat/pull/3719))

- Added keepalive support to `SarvamSTTService` to prevent idle connection
  timeouts (e.g. when used behind a `ServiceSwitcher`).
  (PR [#3730](https://github.com/pipecat-ai/pipecat/pull/3730))

- Added `UserIdleTimeoutUpdateFrame` to enable or disable user idle detection
  at runtime by updating the timeout dynamically.
  (PR [#3748](https://github.com/pipecat-ai/pipecat/pull/3748))

- Added `broadcast_sibling_id` field to the base `Frame` class. This field is
  automatically set by `broadcast_frame()` and `broadcast_frame_instance()` to
  the ID of the paired frame pushed in the opposite direction, allowing
  receivers to identify broadcast pairs.
  (PR [#3774](https://github.com/pipecat-ai/pipecat/pull/3774))

- Added `ignored_sources` parameter to `RTVIObserverParams` and
  `add_ignored_source()`/`remove_ignored_source()` methods to `RTVIObserver` to
  suppress RTVI messages from specific pipeline processors (e.g. a silent
  evaluation LLM).
  (PR [#3779](https://github.com/pipecat-ai/pipecat/pull/3779))

- Added `DeepgramSageMakerTTSService` for running Deepgram TTS models deployed
  on AWS SageMaker endpoints via HTTP/2 bidirectional streaming. Supports the
  Deepgram TTS protocol (Speak, Flush, Clear, Close), interruption handling,
  and per-turn TTFB metrics.
  (PR [#3785](https://github.com/pipecat-ai/pipecat/pull/3785))

### Changed

- ⚠️ `RimeTTSService` now defaults to `model="arcana"` and the
  `wss://users-ws.rime.ai/ws3` endpoint. `InputParams` defaults changed from
  mistv2-specific values to `None` — only explicitly-set params are sent as
  query params.
  (PR [#3642](https://github.com/pipecat-ai/pipecat/pull/3642))

- `AICFilter` now shares read-only AIC models via a singleton `AICModelManager`
  in `aic_filter.py`.
    - Multiple filters using the same model path or `(model_id,
      model_download_dir)` share one loaded model, with reference counting and
      concurrent load deduplication.
    - Model file I/O runs off the event loop so the filter does not block.
  (PR [#3684](https://github.com/pipecat-ai/pipecat/pull/3684))

- Added `X-User-Agent` and `X-Request-Id` headers to `InworldTTSService` for
  better traceability.
  (PR [#3706](https://github.com/pipecat-ai/pipecat/pull/3706))

- `DailyUpdateRemoteParticipantsFrame` is no longer deprecated and is now
  queued with audio like other transport frames.
  (PR [#3719](https://github.com/pipecat-ai/pipecat/pull/3719))

- Bumped Pillow dependency upper bound from `<12` to `<13` to allow Pillow
  12.x.
  (PR [#3728](https://github.com/pipecat-ai/pipecat/pull/3728))

- Moved STT keepalive mechanism from `WebsocketSTTService` to the `STTService`
  base class, allowing any STT service (not just websocket-based ones) to use
  idle-connection keepalive via the `keepalive_timeout` and
  `keepalive_interval` parameters.
  (PR [#3730](https://github.com/pipecat-ai/pipecat/pull/3730))

- Improved audio context management in `AudioContextTTSService` by moving
  context ID tracking to the base class and adding
  `reuse_context_id_within_turn` parameter to control concurrent TTS request
  handling.
    - Added helper methods: `has_active_audio_context()`,
      `get_active_audio_context_id()`, `remove_active_audio_context()`,
      `reset_active_audio_context()`
    - Simplified Cartesia, ElevenLabs, Inworld, Rime, AsyncAI, and Gradium TTS
      implementations by removing duplicate context management code
  (PR [#3732](https://github.com/pipecat-ai/pipecat/pull/3732))

- `UserIdleController` is now always created with a default timeout of 0
  (disabled). The `user_idle_timeout` parameter changed from `Optional[float] =
  None` to `float = 0` in `UserTurnProcessor`, `LLMUserAggregatorParams`, and
  `UserIdleController`.
  (PR [#3748](https://github.com/pipecat-ai/pipecat/pull/3748))

- Change the version specifier from `>=0.2.8` to `~=0.2.8` for the
  `speechmatics-voice` package to ensure compatibility with future patch
  versions.
  (PR [#3761](https://github.com/pipecat-ai/pipecat/pull/3761))

- Updated `InworldTTSService` and `InworldHttpTTSService` to use `ASYNC`
  timestamp transport strategy by default
  (PR [#3765](https://github.com/pipecat-ai/pipecat/pull/3765))

- Added `start_time` and `end_time` parameters to `start_ttfb_metrics()`,
  `stop_ttfb_metrics()`, `start_processing_metrics()`, and
  `stop_processing_metrics()` in `FrameProcessor` and `FrameProcessorMetrics`,
  allowing custom timestamps for metrics measurement. `STTService` now uses
  these instead of custom TTFB tracking.
  (PR [#3776](https://github.com/pipecat-ai/pipecat/pull/3776))

- Updated default Anthropic model from `claude-sonnet-4-5-20250929` to
  `claude-sonnet-4-6`.
  (PR [#3792](https://github.com/pipecat-ai/pipecat/pull/3792))

### Deprecated

- Deprecated unused `Traceable`, `@traceable`, `@traced`, and
  `AttachmentStrategy` in `pipecat.utils.tracing.class_decorators`. This module
  will be removed in a future release.
  (PR [#3733](https://github.com/pipecat-ai/pipecat/pull/3733))

### Fixed

- Fixed race condition where `RTVIObserver` could send messages before
  `DailyTransport` join completed. Outbound messages are now queued & delivered
  after the transport is ready.
  (PR [#3615](https://github.com/pipecat-ai/pipecat/pull/3615))

- Fixed async generator cleanup in OpenAI LLM streaming to prevent
  `AttributeError` with uvloop on Python 3.12+ (MagicStack/uvloop#699).
  (PR [#3698](https://github.com/pipecat-ai/pipecat/pull/3698))

- Fixed `SmallWebRTCTransport` input audio resampling to properly handle all
  sample rates, including 8kHz audio.
  (PR [#3713](https://github.com/pipecat-ai/pipecat/pull/3713))

- Fixed a race condition in `RTVIObserver` where bot output messages could be
  sent before the bot-started-speaking event.
  (PR [#3718](https://github.com/pipecat-ai/pipecat/pull/3718))

- Fixed Grok Realtime `session.updated` event parsing failure caused by the API
  returning prefixed voice names (e.g. `"human_Ara"` instead of `"Ara"`).
  (PR [#3720](https://github.com/pipecat-ai/pipecat/pull/3720))

- Fixed context ID reuse issue in `ElevenLabsTTSService`, `InworldTTSService`,
  `RimeTTSService`, `CartesiaTTSService`, `AsyncAITTSService`, and
  `PlayHTTTSService`. Services now properly reuse the same context ID across
  multiple `run_tts()` invocations within a single LLM turn, preventing context
  tracking issues and incorrect lifecycle signaling.
  (PR [#3729](https://github.com/pipecat-ai/pipecat/pull/3729))

- Fixed word timestamp interleaving issue in `ElevenLabsTTSService` when
  processing multiple sentences within a single LLM turn.
  (PR [#3729](https://github.com/pipecat-ai/pipecat/pull/3729))

- Fixed tracing service decorators executing the wrapped function twice when
  the function itself raised an exception (e.g., LLM rate limit, TTS timeout).
  (PR [#3735](https://github.com/pipecat-ai/pipecat/pull/3735))

- Fixed `LLMUserAggregator` broadcasting mute events before `StartFrame`
  reaches downstream processors.
  (PR [#3737](https://github.com/pipecat-ai/pipecat/pull/3737))

- Fixed `UserIdleController` false idle triggers caused by gaps between user
  and bot activity frames. The idle timer now starts only after
  `BotStoppedSpeakingFrame` and is suppressed during active user turns and
  function calls.
  (PR [#3744](https://github.com/pipecat-ai/pipecat/pull/3744))

- Fixed incorrect `sample_rate` assignment in
  `TavusInputTransport._on_participant_audio_data` (was using
  `audio.audio_frames` instead of `audio.sample_rate`).
  (PR [#3768](https://github.com/pipecat-ai/pipecat/pull/3768))

- Fixed `RTVIObserver` not processing upstream-only frames. Previously, all
  upstream frames were filtered out to avoid duplicate messages from
  broadcasted frames. Now only upstream copies of broadcasted frames are
  skipped.
  (PR [#3774](https://github.com/pipecat-ai/pipecat/pull/3774))

- Fixed mutable default arguments in `LLMContextAggregatorPair.__init__()` that
  could cause shared state across instances.
  (PR [#3782](https://github.com/pipecat-ai/pipecat/pull/3782))

- Fixed `DeepgramSageMakerSTTService` to properly track finalize lifecycle
  using `request_finalize()` / `confirm_finalize()` and use `is_final` (instead
  of `is_final and speech_final`) for final transcription detection, matching
  `DeepgramSTTService` behavior.
  (PR [#3784](https://github.com/pipecat-ai/pipecat/pull/3784))

- Fixed a race condition in `AudioContextTTSService` where the audio context
  could time out between consecutive TTS requests within the same turn, causing
  audio to be discarded.
  (PR [#3787](https://github.com/pipecat-ai/pipecat/pull/3787))

- Fixed `push_interruption_task_frame_and_wait()` hanging indefinitely when the
  `InterruptionFrame` does not reach the pipeline sink within the timeout.
  Added a `timeout` keyword argument to customize the wait duration.
  (PR [#3789](https://github.com/pipecat-ai/pipecat/pull/3789))

## [0.0.102] - 2026-02-10

### Added

- Added `ResembleAITTSService` for text-to-speech using Resemble AI's streaming
  WebSocket API with word-level timestamps and jitter buffering for smooth
  audio playback.
  (PR [#3134](https://github.com/pipecat-ai/pipecat/pull/3134))

- Added `UserBotLatencyObserver` for tracking user-to-bot response latency.
  When tracing is enabled, latency measurements are automatically recorded as
  `turn.user_bot_latency_seconds` attributes on OpenTelemetry turn spans.
  (PR [#3355](https://github.com/pipecat-ai/pipecat/pull/3355))

- Added `append_to_context` parameter to `TTSSpeakFrame` for conditional LLM
  context addition.
    - Allows fine-grained control over whether text should be added to
      conversation context
    - Defaults to `True` to maintain backward compatibility
  (PR [#3584](https://github.com/pipecat-ai/pipecat/pull/3584))

- Added TTS context tracking system with `context_id` field to trace audio
  generation through the pipeline.
    - `TTSAudioRawFrame`, `TTSStartedFrame`, `TTSStoppedFrame` now include
      `context_id`
    - `AggregatedTextFrame` and `TTSTextFrame` now include `context_id`
    - Enables tracking which TTS request generated specific audio chunks
  (PR [#3584](https://github.com/pipecat-ai/pipecat/pull/3584))

- Added support for Inworld TTS Websocket Auto Mode for improved latency
  (PR [#3593](https://github.com/pipecat-ai/pipecat/pull/3593))

- Added new frames for context summarization: `LLMContextSummaryRequestFrame`
  and `LLMContextSummaryResultFrame`.
  (PR [#3621](https://github.com/pipecat-ai/pipecat/pull/3621))

- Added context summarization feature to automatically compress conversation
  history when conversation length limits (by token or message count) are
  reached, enabling efficient long-running conversations.
    - Configure via `enable_context_summarization=True` in
      `LLMAssistantAggregatorParams`
    - Customize behavior with `LLMContextSummarizationConfig` (max tokens,
      thresholds, etc.)
    - Automatically preserves incomplete function call sequences during
      summarization
    - See new examples:
  `examples/foundational/54-context-summarization-openai.py` and
  `examples/foundational/54a-context-summarization-google.py`
  (PR [#3621](https://github.com/pipecat-ai/pipecat/pull/3621))

- Added RTVI function call lifecycle events (`llm-function-call-started`,
  `llm-function-call-in-progress`, `llm-function-call-stopped`) with
  configurable security levels via
  `RTVIObserverParams.function_call_report_level`. Supports per-function
  control over what information is exposed (`DISABLED`, `NONE`, `NAME`, or
  `FULL`).
  (PR [#3630](https://github.com/pipecat-ai/pipecat/pull/3630))

- Added `RequestMetadataFrame` and metadata handling for `ServiceSwitcher` to
  ensure STT services correctly emit `STTMetadataFrame` when switching between
  services. Only the active service's metadata is propagated downstream,
  switching services triggers the newly active service to re-emit its metadata,
  and proper frame ordering is maintained at startup.
  (PR [#3637](https://github.com/pipecat-ai/pipecat/pull/3637))

- Added `STTMetadataFrame` to broadcast STT service latency information at
  pipeline start.
    - STT services broadcast P99 time-to-final-segment (`ttfs_p99_latency`) to
      downstream processors
    - Turn stop strategies automatically configure their STT timeout from this
      metadata
    - Developers can override `ttfs_p99_latency` via constructor argument for
      custom deployments
    - Added measured P99 values for STT providers.
    - See [stt-benchmark](https://github.com/pipecat-ai/stt-benchmark) to
      measure latency for your configuration
  (PR [#3637](https://github.com/pipecat-ai/pipecat/pull/3637))

- Added support for `is_sandbox` parameter in `LiveAvatarNewSessionRequest` to
  enable sandbox mode for HeyGen LiveAvatar sessions.
  (PR [#3653](https://github.com/pipecat-ai/pipecat/pull/3653))

- Added support for `video_settings` parameter in `LiveAvatarNewSessionRequest`
  to configure video encoding (H264/VP8) and quality levels.
  (PR [#3653](https://github.com/pipecat-ai/pipecat/pull/3653))

- Added `OpenAIRealtimeSTTService` for real-time streaming speech-to-text using
  OpenAI's Realtime API WebSocket transcription sessions. Supports local VAD
  and server-side VAD modes, noise reduction, and automatic reconnection.
  (PR [#3656](https://github.com/pipecat-ai/pipecat/pull/3656))

- Added `bulbul:v3-beta` TTS model support for Sarvam AI with temperature
  control and 25 new speaker voices.
  (PR [#3671](https://github.com/pipecat-ai/pipecat/pull/3671))

- Added `saaras:v3` STT model support for Sarvam AI with new `mode` parameter
  (transcribe, translate, verbatim, translit, codemix) and prompt support.
  (PR [#3671](https://github.com/pipecat-ai/pipecat/pull/3671))

- Added new OpenAI TTS voice options `marin` and `cedar`.
  (PR [#3682](https://github.com/pipecat-ai/pipecat/pull/3682))

- Added `UserMuteStartedFrame` and `UserMuteStoppedFrame` system frames, and
  corresponding `user-mute-started` / `user-mute-stopped` RTVI messages, so
  clients can observe when mute strategies activate or deactivate.
  (PR [#3687](https://github.com/pipecat-ai/pipecat/pull/3687))

### Changed

- Updated all 30+ TTS service implementations to support context tracking with
  `context_id`.
    - Services now generate and propagate context IDs through TTS frames
    - Enables end-to-end tracing of TTS requests through the pipeline
  (PR [#3584](https://github.com/pipecat-ai/pipecat/pull/3584))

- ⚠️ `TTSService.run_tts()` now requires a `context_id` parameter for context
  tracking.
    - Custom TTS service implementations must update their `run_tts()`
      signature
    - Before: `async def run_tts(self, text: str) -> AsyncGenerator[Frame,
      None]:`
    - After: `async def run_tts(self, text: str, context_id: str) ->
      AsyncGenerator[Frame, None]:`
  (PR [#3584](https://github.com/pipecat-ai/pipecat/pull/3584))

- Simplified context aggregators to use `frame.append_to_context` flag instead
  of tracking internal state.
    - Cleaner logic in `LLMResponseAggregator` and
      `LLMResponseUniversalAggregator`
    - More consistent behavior across aggregator implementations
  (PR [#3584](https://github.com/pipecat-ai/pipecat/pull/3584))

- Updated timestamps to be cumulative within an agent turn, using
  flushCompleted message as an indication of when timestamps from the server
  are reset to 0
  (PR [#3593](https://github.com/pipecat-ai/pipecat/pull/3593))

- Changed `KokoroTTSService` to use `kokoro-onnx` instead of `kokoro` as the
  underlying TTS engine.
  (PR [#3612](https://github.com/pipecat-ai/pipecat/pull/3612))

- Improved user turn stop timing in `TranscriptionUserTurnStopStrategy` and
  `TurnAnalyzerUserTurnStopStrategy`.
    - Timeout now starts on `VADUserStoppedSpeakingFrame` for tighter, more
      predictable timing
    - Added support for finalized transcripts
      (`TranscriptionFrame.finalized=True`) to trigger earlier
    - Added fallback timeout for edge cases where transcripts arrive without
      VAD events
    - Removed `InterimTranscriptionFrame` handling (no longer affects timing)
  (PR [#3637](https://github.com/pipecat-ai/pipecat/pull/3637))

- Improved the accuracy of the `UserBotLatencyObserver` and
  `UserBotLatencyLogObserver` by measuring from the time when the user actually
  starts speaking.
  (PR [#3637](https://github.com/pipecat-ai/pipecat/pull/3637))

- ⚠️ Renamed `timeout` parameter to `user_speech_timeout` in
  `TranscriptionUserTurnStopStrategy`.
  (PR [#3637](https://github.com/pipecat-ai/pipecat/pull/3637))

- Updated the `VADUserStartedSpeakingFrame` to include `start_secs` and
  `timestamp` and `VADUserStoppedSpeakingFrame` to include `stop_secs` and
  `timestamp`, removing the need to separately handle the
  `SpeechControlParamsFrame` for VADParams values.
  (PR [#3637](https://github.com/pipecat-ai/pipecat/pull/3637))

- ⚠️ Renamed `TranscriptionUserTurnStopStrategy` to
  `SpeechTimeoutUserTurnStopStrategy`. The old name is deprecated and will be
  removed in a future release.
  (PR [#3637](https://github.com/pipecat-ai/pipecat/pull/3637))

- `AssemblyAISTTService` now automatically configures optimal settings for
  manual turn detection when `vad_force_turn_endpoint=True`. This sets
  `end_of_turn_confidence_threshold=1.0` and `max_turn_silence=2000` by
  default, which disables model-based turn detection and reduces latency by
  relying on external VAD for turn endpoints. Warnings are logged if
  conflicting settings are detected.
  (PR [#3644](https://github.com/pipecat-ai/pipecat/pull/3644))

- Upgraded the `pipecat-ai-small-webrtc-prebuilt` package to v2.1.0.
  (PR [#3652](https://github.com/pipecat-ai/pipecat/pull/3652))

- Changed default session mode from "CUSTOM" to "LITE" in HeyGen LiveAvatar
  integration, with VP8 as the default video encoding.
  (PR [#3653](https://github.com/pipecat-ai/pipecat/pull/3653))

- ⚠️ The default `VADParams` `stop_secs` default is changing from `0.8` seconds
  to `0.2` seconds. This change both simplifies the developer experience and
  improves the performance of STT services. With a shorter `stop_secs` value,
  STT services using a local VAD can finalize sooner, resulting in faster
  transcription.
    - `SpeechTimeoutUserTurnStopStrategy`: control how long to wait for
      additional user speech using `user_speech_timeout` (default: 0.6 sec).
    - `TurnAnalyzerUserTurnStopStrategy`: the turn analyzer automatically
      adjusts the user wait time based on the audio input.
  (PR [#3659](https://github.com/pipecat-ai/pipecat/pull/3659))

- Moved interruption wait event from per-processor instance state to
  `InterruptionFrame` itself. Added `InterruptionFrame.complete()` to signal
  when the interruption has fully traversed the pipeline. Custom processors
  that block or consume an `InterruptionFrame` before it reaches the pipeline
  sink must call `frame.complete()` to avoid stalling
  `push_interruption_task_frame_and_wait()`. A warning is logged if completion
  does not happen within 2 seconds.
  (PR [#3660](https://github.com/pipecat-ai/pipecat/pull/3660))

- Update the default model to `scribe_v2` for `ElevenLabsSTTService`.
  (PR [#3664](https://github.com/pipecat-ai/pipecat/pull/3664))

- Changed the `DeepgramSTTService` default setting for `smart_format` to
  `False`, as agents don't need smart formatting. Disabling this setting
  provides a small performance improvement, as well.
  (PR [#3666](https://github.com/pipecat-ai/pipecat/pull/3666))

- Changed `FunctionCallCancelFrame` to broadcast in both directions for
  consistency with other function call frames.
  (PR [#3672](https://github.com/pipecat-ai/pipecat/pull/3672))

- Changed default user turn stop strategy from
  `TranscriptionUserTurnStopStrategy` to `TurnAnalyzerUserTurnStopStrategy`
  with `LocalSmartTurnAnalyzerV3`.
  (PR [#3689](https://github.com/pipecat-ai/pipecat/pull/3689))

- Renamed `RequestMetadataFrame` to `ServiceSwitcherRequestMetadataFrame` and
  added a `service` field to target a specific service. The frame is now pushed
  downstream by services after handling instead of being silently consumed.
  (PR [#3692](https://github.com/pipecat-ai/pipecat/pull/3692))

- Update `SonioxSTTService` to set `vad_force_turn_endpoint` to `True`. This
  setting disabled the turn detection logic available natively in Soniox.
  Instead, Soniox relies on a local VAD to finalize the transcript. This
  configuration meaningfully reduces the time to final segment for Soniox. With
  this setting enabled, Soniox outputs a transcript in ~250ms (median). Pipecat
  enables smart-turn detection by default using the `LocalSmartTurnAnalyzerV3`.
  To use the native turn detection logic in Soniox, just set
  `vad_force_turn_endpoint` to `False`.
  (PR [#3697](https://github.com/pipecat-ai/pipecat/pull/3697))

- Update `SonioxSTTService` default model to `stt-rt-v4`.
  (PR [#3697](https://github.com/pipecat-ai/pipecat/pull/3697))

- Updated the default model to `async_flash_v1.0` and base URL to
  `https://api.async.com` for `AsyncAITTSService`.
  (PR [#3701](https://github.com/pipecat-ai/pipecat/pull/3701))

### Deprecated

- Deprecated `UserBotLatencyLogObserver`. Use `UserBotLatencyObserver` directly
  with its `on_latency_measured` event handler instead.
  (PR [#3355](https://github.com/pipecat-ai/pipecat/pull/3355))

- Deprecated `RTVILLMFunctionCallMessage`, `RTVILLMFunctionCallMessageData`,
  and `RTVIProcessor.handle_function_call()`. Use the new
  `llm-function-call-in-progress` event sent automatically by `RTVIObserver`
  instead.
  (PR [#3630](https://github.com/pipecat-ai/pipecat/pull/3630))

### Removed

- ⚠️ Removed `timeout` parameter from `TurnAnalyzerUserTurnStopStrategy`. The
  timeout is now managed internally based on STT latency.
  (PR [#3637](https://github.com/pipecat-ai/pipecat/pull/3637))

### Fixed

- Fixed pipeline freeze when `InterruptionFrame` discards `EndFrame` or
  `StopFrame` by making terminal frames uninterruptible.
  (PR [#3542](https://github.com/pipecat-ai/pipecat/pull/3542))

- Fixed OpenAI LLM stream not being closed on cancellation/exception, which
  could leak sockets.
  (PR [#3589](https://github.com/pipecat-ai/pipecat/pull/3589))

- Fixed `PipelineTask` adding duplicate `RTVIProcessor` and `RTVIObserver` when
  they were already provided in the pipeline or observers list. They are now
  detected and skipped, with appropriate warnings and errors logged for
  mismatched configurations.
  (PR [#3610](https://github.com/pipecat-ai/pipecat/pull/3610))

- Fixed function call timeout task not being cancelled when the handler
  completes without calling `result_callback` or is cancelled externally, which
  caused `RuntimeWarning: coroutine was never awaited`.
  (PR [#3616](https://github.com/pipecat-ai/pipecat/pull/3616))

- Fixed sentence splitting for Japanese, Chinese, Korean, and other non-Latin
  languages in TTS pipeline. NLTK's sentence tokenizer does not support CJK
  languages, causing text to accumulate until flush instead of being split at
  sentence boundaries. Added fallback detection for unambiguous non-Latin
  sentence-ending punctuation (e.g., `。`, `？`, `！`).
  (PR [#3617](https://github.com/pipecat-ai/pipecat/pull/3617))

- Fixed `PipelineTask` to also call `set_bot_ready()` when an external
  `RTVIProcessor` is provided.
  (PR [#3623](https://github.com/pipecat-ai/pipecat/pull/3623))

- Fixed `VADController` not broadcasting `SpeechControlParamsFrame` on startup,
  which prevented STT services from receiving VAD params needed for TTFB
  measurement.
  (PR [#3628](https://github.com/pipecat-ai/pipecat/pull/3628))

- Fixed `StopAsyncIteration` exceptions in `parse_telephony_websocket()` when
  WebSocket connections close before sending expected messages.
  (PR [#3629](https://github.com/pipecat-ai/pipecat/pull/3629))

- Fixed WebSocket transport error when broadcasting
  `InputTransportMessageFrame` by correctly instantiating the frame with its
  message parameter.
  (PR [#3635](https://github.com/pipecat-ai/pipecat/pull/3635))

- Fixed orphan OpenTelemetry spans during flow initialization and transitions
  in tracing.
  (PR [#3649](https://github.com/pipecat-ai/pipecat/pull/3649))

- Fixed `SambaNovaLLMService` and `GoogleLLMOpenAIBetaService` streams not
  being closed on cancellation/exception, which could leak sockets.
  (PR [#3663](https://github.com/pipecat-ai/pipecat/pull/3663))

- Fixed an issue in `InworldTTSService` where punctuation was pronounced. Now,
  the `InworldTTSService` ensures proper spacing between sentences, resolving
  pronunciation issues.
  (PR [#3667](https://github.com/pipecat-ai/pipecat/pull/3667))

- Fixed `ParallelPipeline` allowing frames pushed by internal processors to
  escape during lifecycle frame (`StartFrame`/`EndFrame`/`CancelFrame`)
  synchronization. These frames are now buffered and flushed after all branches
  complete.
  (PR [#3668](https://github.com/pipecat-ai/pipecat/pull/3668))

- Fixed issues in Sarvam STT and TTS services: missing event handler
  registration for VAD signals, `Optional[bool]` type annotations, WebSocket
  state cleanup on API errors, and TTS disconnect/reconnection state
  management.
  (PR [#3671](https://github.com/pipecat-ai/pipecat/pull/3671))

- Fixed `RTVIObserver` sending duplicate client messages for frames that are
  broadcast in both directions (e.g. `UserStartedSpeakingFrame`,
  `FunctionCallResultFrame`).
  (PR [#3672](https://github.com/pipecat-ai/pipecat/pull/3672))

- Fixed WebSocket STT services (ElevenLabs, Cartesia, Gladia, Soniox)
  disconnecting due to idle timeout when no audio is being sent (e.g. when
  inactive behind a `ServiceSwitcher`). `WebsocketSTTService` now provides
  opt-in silence-based keepalive via `keepalive_timeout` and
  `keepalive_interval` parameters.
  (PR [#3675](https://github.com/pipecat-ai/pipecat/pull/3675))

## [0.0.101] - 2026-01-30

### Added

- Additions for `AICFilter` and `AICVADAnalyzer`:
    - Added model downloading support to `AICFilter` with `model_id` and
      `model_download_dir` parameters.
    - Added `model_path` parameter to `AICFilter` for loading local `.aicmodel`
      files.
    - Added unit tests for `AICFilter` and `AICVADAnalyzer`.
  (PR [#3408](https://github.com/pipecat-ai/pipecat/pull/3408))

- Added handling for `server_content.interrupted` signal in the Gemini Live
  service for faster interruption response in the case where there isn't
  already turn tracking in the pipeline, e.g. local VAD + context aggregators.
  When there is already turn tracking in the pipeline, the additional
  interruption does no harm.
  (PR [#3429](https://github.com/pipecat-ai/pipecat/pull/3429))

- Added new `GenesysFrameSerializer` for the Genesys AudioHook WebSocket
  protocol, enabling bidirectional audio streaming between Pipecat pipelines
  and Genesys Cloud contact center.
  (PR [#3500](https://github.com/pipecat-ai/pipecat/pull/3500))

- Added `reached_upstream_types` and `reached_downstream_types` read-only
  properties to `PipelineTask` for inspecting current frame filters.
  (PR [#3510](https://github.com/pipecat-ai/pipecat/pull/3510))

- Added `add_reached_upstream_filter()` and `add_reached_downstream_filter()`
  methods to `PipelineTask` for appending frame types.
  (PR [#3510](https://github.com/pipecat-ai/pipecat/pull/3510))

- Added `UserTurnCompletionLLMServiceMixin` for LLM services to detect and
  filter incomplete user turns. When enabled via `filter_incomplete_user_turns`
  in `LLMUserAggregatorParams`, the LLM outputs a turn completion marker at the
  start of each response: ✓ (complete), ○ (incomplete short), or ◐ (incomplete
  long). Incomplete turns are suppressed, and configurable timeouts
  automatically re-prompt the user.
  (PR [#3518](https://github.com/pipecat-ai/pipecat/pull/3518))

- Added `FrameProcessor.broadcast_frame_instance(frame)` method to broadcast a
  frame instance by extracting its fields and creating new instances for each
  direction.
  (PR [#3519](https://github.com/pipecat-ai/pipecat/pull/3519))

- `PipelineTask` now automatically adds `RTVIProcessor` and registers
  `RTVIObserver` when `enable_rtvi=True` (default), simplifying pipeline setup.
  (PR [#3519](https://github.com/pipecat-ai/pipecat/pull/3519))

- Added `RTVIProcessor.create_rtvi_observer()` factory method for creating RTVI
  observers.
  (PR [#3519](https://github.com/pipecat-ai/pipecat/pull/3519))

- Added `video_out_codec` parameter to `TransportParams` allowing configuration
  of the preferred video codec (e.g., `"VP8"`, `"H264"`, `"H265"`) for video
  output in `DailyTransport`.
  (PR [#3520](https://github.com/pipecat-ai/pipecat/pull/3520))

- Added `location` parameter to Google TTS services (`GoogleHttpTTSService`,
  `GoogleTTSService`, `GeminiTTSService`) for regional endpoint support.
  (PR [#3523](https://github.com/pipecat-ai/pipecat/pull/3523))

- Added new `PIPECAT_SMART_TURN_LOG_DATA` environment variable, which causes
  Smart Turn input data to be saved to disk
  (PR [#3525](https://github.com/pipecat-ai/pipecat/pull/3525))

- Added `result_callback` parameter to `UserImageRequestFrame` to support
  deferred function call results.
  (PR [#3571](https://github.com/pipecat-ai/pipecat/pull/3571))

- Added `function_call_timeout_secs` parameter to `LLMService` to configure
  timeout for deferred function calls (defaults to 10.0 seconds).
  (PR [#3571](https://github.com/pipecat-ai/pipecat/pull/3571))

- Added `vad_analyzer` parameter to `LLMUserAggregatorParams`. VAD analysis is
  now handled inside the `LLMUserAggregator` rather than in the transport,
  keeping voice activity detection closer to where it is consumed. The
  `vad_analyzer` on `BaseInputTransport` is now deprecated.

    ```python
    context_aggregator = LLMContextAggregatorPair(
        context,
        user_params=LLMUserAggregatorParams(
            vad_analyzer=SileroVADAnalyzer(),
        ),
    )
    ```
  (PR [#3583](https://github.com/pipecat-ai/pipecat/pull/3583))

- Added `VADProcessor` for detecting speech in audio streams within a pipeline.
  Pushes `VADUserStartedSpeakingFrame`, `VADUserStoppedSpeakingFrame`, and
  `UserSpeakingFrame` downstream based on VAD state changes.
  (PR [#3583](https://github.com/pipecat-ai/pipecat/pull/3583))

- Added `VADController` for managing voice activity detection state and
  emitting speech events independently of transport or pipeline processors.
  (PR [#3583](https://github.com/pipecat-ai/pipecat/pull/3583))

- Added local `PiperTTSService` for offline text-to-speech using Piper voice
  models. The existing HTTP-based service has been renamed to
  `PiperHttpTTSService`.
  (PR [#3585](https://github.com/pipecat-ai/pipecat/pull/3585))

- `main()` in `pipecat.runner.run` now accepts an optional
  `argparse.ArgumentParser`, allowing bots to define custom CLI arguments
  accessible via `runner_args.cli_args`.
  (PR [#3590](https://github.com/pipecat-ai/pipecat/pull/3590))

- Added `KokoroTTSService` for local text-to-speech synthesis using the
  Kokoro-82M model.
  (PR [#3595](https://github.com/pipecat-ai/pipecat/pull/3595))

### Changed

- Updated `AICFilter` and `AICVADAnalyzer` to use aic-sdk ~= 2.0.1.
  (PR [#3408](https://github.com/pipecat-ai/pipecat/pull/3408))

- Improved the STT TTFB (Time To First Byte) measurement, reporting the delay
  between when the user stops speaking and when the final transcription is
  received. Note: Unlike traditional TTFB which measures from a discrete
  request, STT services receive continuous audio input—so we measure from
  speech end to final transcript, which captures the latency that matters for
  voice AI applications. In support of this change, added `finalized` field to
  `TranscriptionFrame` to indicate when a transcript is the final result for an
  utterance.
  (PR [#3495](https://github.com/pipecat-ai/pipecat/pull/3495))

- `SarvamSTTService` now defaults `vad_signals` and `high_vad_sensitivity` to
  `None` (omitted from connection parameters), improving latency by ~300ms
  compared to the previous defaults.
  (PR [#3495](https://github.com/pipecat-ai/pipecat/pull/3495))

- Changed frame filter storage from tuples to sets in `PipelineTask`.
  (PR [#3510](https://github.com/pipecat-ai/pipecat/pull/3510))

- Changed default Inworld TTS model from `inworld-tts-1` to
  `inworld-tts-1.5-max`.
  (PR [#3531](https://github.com/pipecat-ai/pipecat/pull/3531))

- `FrameSerializer` now subclasses from `BaseObject` to enable event support.
  (PR [#3560](https://github.com/pipecat-ai/pipecat/pull/3560))

- Added support for TTFS in `SpeechmaticsSTTService` and set the default mode
  to `EXTERNAL` to support Pipecat-controlled VAD.
  - Changed dependency to `speechmatics-voice[smart]>=0.2.8`
  (PR [#3562](https://github.com/pipecat-ai/pipecat/pull/3562))

- ⚠️ Changed function call handling to use timeout-based completion instead of
  immediate callback execution.
    - Function calls that defer their results (e.g., `UserImageRequestFrame`)
      now use a timeout mechanism
    - The `result_callback` is invoked automatically when the deferred
      operation completes or after timeout
    - This change affects examples using `UserImageRequestFrame` - the
      `result_callback` should now be passed to the frame instead of being called
      immediately
  (PR [#3571](https://github.com/pipecat-ai/pipecat/pull/3571))

- Pipecat runner now uses `DAILY_ROOM_URL` instead of `DAILY_SAMPLE_ROOM_URL`.
  (PR [#3582](https://github.com/pipecat-ai/pipecat/pull/3582))

- Updates to `GradiumSTTService`:
    - Now flushes pending transcriptions when VAD detects the user stopped
      speaking, improving response latency.
    - `GradiumSTTService` now supports `InputParams` for configuring `language`
      and `delay_in_frames` settings.
  (PR [#3587](https://github.com/pipecat-ai/pipecat/pull/3587))

### Deprecated

- ⚠️ Deprecated `vad_analyzer` parameter on `BaseInputTransport`. Pass
  `vad_analyzer` to `LLMUserAggregatorParams` instead or use `VADProcessor` in
  the pipeline.
  (PR [#3583](https://github.com/pipecat-ai/pipecat/pull/3583))

### Removed

- Removed deprecated `AICFilter` parameters: `enhancement_level`, `voice_gain`,
  `noise_gate_enable`.
  (PR [#3408](https://github.com/pipecat-ai/pipecat/pull/3408))

### Fixed

- Fixed an issue where if you were using `OpenRouterLLMService` with a Gemini
  model, it wouldn't handle multiple `"system"` messages as expected (and as we
  do in `GoogleLLMService`), which is to convert subsequent ones into `"user"`
  messages. Instead, the latest `"system"` message would overwrite the previous
  ones.
  (PR [#3406](https://github.com/pipecat-ai/pipecat/pull/3406))

- Transports now properly broadcast `InputTransportMessageFrame` frames both
  upstream and downstream instead of only pushing downstream.
  (PR [#3519](https://github.com/pipecat-ai/pipecat/pull/3519))

- Fixed `FrameProcessor.broadcast_frame()` to deep copy kwargs, preventing
  shared mutable references between the downstream and upstream frame
  instances.
  (PR [#3519](https://github.com/pipecat-ai/pipecat/pull/3519))

- Fixed OpenAI LLM services to emit `ErrorFrame` on completion timeout,
  enabling proper error handling and LLMSwitcher failover.
  (PR [#3529](https://github.com/pipecat-ai/pipecat/pull/3529))

- Fixed a logging issue where non-ASCII characters (e.g., Japanese, Chinese,
  etc.) were being unnecessarily escaped to Unicode sequences when function
  call occurred.
  (PR [#3536](https://github.com/pipecat-ai/pipecat/pull/3536))

- Fixed how audio tracks are synchronized inside the `AudioBufferProcessor` to
  fix timing issues where silence and audio were misaligned between user and
  bot buffers.
  (PR [#3541](https://github.com/pipecat-ai/pipecat/pull/3541))

- Fixed race condition in `OpenAIRealtimeBetaLLMService` that could cause an
  error when truncating the conversation.
  (PR [#3567](https://github.com/pipecat-ai/pipecat/pull/3567))

- Fixed an infinite loop in `WebsocketService` that blocked the event loop when
  a remote server closed the connection gracefully.
  (PR [#3574](https://github.com/pipecat-ai/pipecat/pull/3574))

- Fixed `LLMUserAggregator` and `LLMAssistantAggregator` not emitting pending
  transcripts via `on_user_turn_stopped` and `on_assistant_turn_stopped` events
  when the conversation ends (`EndFrame`) or is cancelled (`CancelFrame`).
  (PR [#3575](https://github.com/pipecat-ai/pipecat/pull/3575))

- Added missing `LiveKitRunnerArguments` and `LiveKitTransport` support in
  runner utilities to enable LiveKit transport configuration.
  (PR [#3580](https://github.com/pipecat-ai/pipecat/pull/3580))

- Fixed race condition in `OpenAIRealtimeLLMService` that could cause an error
  when truncating the conversation.
  (PR [#3581](https://github.com/pipecat-ai/pipecat/pull/3581))

- Fixed `PiperHttpTTSService` (olf `PiperTTSService`) to resample audio output
  based on the model's sample rate parsed from the WAV header.
  (PR [#3585](https://github.com/pipecat-ai/pipecat/pull/3585))

- Fixed `UserTurnController` to reset user turn timeout when interim
  transcriptions are received.
  (PR [#3594](https://github.com/pipecat-ai/pipecat/pull/3594))

- Fixed an issue in the `IVRNavigator` where the `TextFrame`s pushed had
  incorrect spacing. Now, the internal `IVRProcessor` pushes
  `AggregatedTextFrame`s when in conversation mode. This allows for controlling
  spacing of the outputted, aggregated text.
  (PR [#3604](https://github.com/pipecat-ai/pipecat/pull/3604))

- Fixed `GeminiLiveLLMService` transcription timeout handler not being
  scheduled by yielding to the event loop after task creation.
  (PR [#3605](https://github.com/pipecat-ai/pipecat/pull/3605))

## [0.0.100] - 2026-01-20

### Added

- Added Hathora service to support Hathora-hosted TTS and STT models (only
  non-streaming)
  (PR [#3169](https://github.com/pipecat-ai/pipecat/pull/3169))

- Added `CambTTSService`, using Camb.ai's TTS integration with MARS models
  (mars-flash, mars-pro, mars-instruct) for high-quality text-to-speech
  synthesis.
  (PR [#3349](https://github.com/pipecat-ai/pipecat/pull/3349))

- Added the `additional_headers` param to `WebsocketClientParams`, allowing
  `WebsocketClientTransport` to send custom headers on connect, for cases such
  as authentication.
  (PR [#3461](https://github.com/pipecat-ai/pipecat/pull/3461))

- Added `UserIdleController` for detecting user idle state, integrated into
  `LLMUserAggregator` and `UserTurnProcessor` via optional `user_idle_timeout`
  parameter. Emits `on_user_turn_idle` event for application-level handling.
  Deprecated `UserIdleProcessor` in favor of the new compositional approach.
  (PR [#3482](https://github.com/pipecat-ai/pipecat/pull/3482))

- Added `on_user_mute_started` and `on_user_mute_stopped` event handlers to
  `LLMUserAggregator` for tracking user mute state changes.
  (PR [#3490](https://github.com/pipecat-ai/pipecat/pull/3490))

### Changed

- Enhanced interruption handling in `AsyncAITTSService` by supporting
  multi-context WebSocket sessions for more robust context management.
  (PR [#3287](https://github.com/pipecat-ai/pipecat/pull/3287))

- Throttle `UserSpeakingFrame` to broadcast at most every 200ms instead of on
  every audio chunk, reducing frame processing overhead during user speech.
  (PR [#3483](https://github.com/pipecat-ai/pipecat/pull/3483))

### Deprecated

- For consistency with other package names, we just deprecated
  `pipecat.turns.mute` (introduced in Pipecat 0.0.99) in favor of
  `pipecat.turns.user_mute`.
  (PR [#3479](https://github.com/pipecat-ai/pipecat/pull/3479))

### Fixed

- Corrected TTFB metric calculation in `AsyncAIHttpTTSService`.
  (PR [#3287](https://github.com/pipecat-ai/pipecat/pull/3287))

- Fixed an issue where the "bot-llm-text" RTVI event would not fire for
  realtime (speech-to-speech) services:

    - `AWSNovaSonicLLMService`
    - `GeminiLiveLLMService`
    - `OpenAIRealtimeLLMService`
    - `GrokRealtimeLLMService`

  The issue was that these services weren't pushing `LLMTextFrame`s. Now
  they  do.
  (PR [#3446](https://github.com/pipecat-ai/pipecat/pull/3446))

- Fixed an issue where `on_user_turn_stop_timeout` could fire while a user is
  talking when using `ExternalUserTurnStrategies`.
  (PR [#3454](https://github.com/pipecat-ai/pipecat/pull/3454))

- Fixed an issue where user turn start strategies were not being reset after a
  user turn started, causing incorrect strategy behavior.
  (PR [#3455](https://github.com/pipecat-ai/pipecat/pull/3455))

- Fixed `MinWordsUserTurnStartStrategy` to not aggregate transcriptions,
  preventing incorrect turn starts when words are spoken with pauses between
  them.
  (PR [#3462](https://github.com/pipecat-ai/pipecat/pull/3462))

- Fixed an issue where Grok Realtime would error out when running with
  SmallWebRTC transport.
  (PR [#3480](https://github.com/pipecat-ai/pipecat/pull/3480))

- Fixed a `Mem0MemoryService` issue where passing `async_mode: true` was
  causing an error. See
  https://docs.mem0.ai/platform/features/async-mode-default-change.
  (PR [#3484](https://github.com/pipecat-ai/pipecat/pull/3484))

- Fixed `AWSNovaSonicLLMService.reset_conversation()`, which would previously
  error out. Now it successfully reconnects and "rehydrates" from the context
  object.
  (PR [#3486](https://github.com/pipecat-ai/pipecat/pull/3486))

- Fixed `AzureTTSService` transcript formatting issues:
    - Punctuation now appears without extra spaces (e.g., "Hello!" instead of
      "Hello !")
    - CJK languages (Chinese, Japanese, Korean) no longer have unwanted spaces
      between characters
  (PR [#3489](https://github.com/pipecat-ai/pipecat/pull/3489))

- Fixed an issue where `UninterruptibleFrame` frames would not be preserved in
  some cases.
  (PR [#3494](https://github.com/pipecat-ai/pipecat/pull/3494))

- Fixed memory leak in `LiveKitTransport` when `video_in_enabled` is `False`.
  (PR [#3499](https://github.com/pipecat-ai/pipecat/pull/3499))

- Fixed an issue in `AIService` where unhandled exceptions in `start()`,
  `stop()`, or `cancel()` implementations would prevent `process_frame()` to
  continue and therefore `StartFrame`, `EndFrame`, or `CancelFrame` from being
  pushed downstream, causing the pipeline to not start or stop properly.
  (PR [#3503](https://github.com/pipecat-ai/pipecat/pull/3503))

- Moved `NVIDIATTSService` and `NVIDIASTTService` client initialization from
  constructor to `start()` for better error handling.
  (PR [#3504](https://github.com/pipecat-ai/pipecat/pull/3504))

- Optimized `NVIDIATTSService` to process incoming audio frames immediately.
  (PR [#3509](https://github.com/pipecat-ai/pipecat/pull/3509))

- Optimized `NVIDIASTTService` by removing unnecessary queue and task.
  (PR [#3509](https://github.com/pipecat-ai/pipecat/pull/3509))

- Fixed a `CambTTSService` issue where client was being initialized in the
  constructor which wouldn't allow for proper Pipeline error handling.
  (PR [#3511](https://github.com/pipecat-ai/pipecat/pull/3511))

## [0.0.99] - 2026-01-13

### Added

- Introducing user turn strategies. User turn strategies indicate when the user
  turn starts or stops. In conversational agents, these are often referred to
  as start/stop speaking or turn-taking plans or policies.

  User turn start strategies indicate when the user starts speaking (e.g.
  using VAD events or when a user says one or more words).

  User turn stop strategies indicate when the user stops speaking (e.g. using
  an end-of-turn detection model or by observing incoming transcriptions).

  A list of strategies can be specified for both strategies; strategies are
  evaluated in order until one evaluates to true.

  Available user turn start strategies:

  - VADUserTurnStartStrategy
  - TranscriptionUserTurnStartStrategy
  - MinWordsUserTurnStartStrategy
  - ExternalUserTurnStartStrategy

  Available user turn stop strategies:

  - TranscriptionUserTurnStopStrategy
  - TurnAnalyzerUserTurnStopStrategy
  - ExternalUserTurnStopStrategy

  The default strategies are:

  - start: [VADUserTurnStartStrategy, TranscriptionUserTurnStartStrategy]
  - stop: [TranscriptionUserTurnStopStrategy]

  Turn strategies are configured when setting up `LLMContextAggregatorPair`.
  For example:

  ```python
  context_aggregator = LLMContextAggregatorPair(
      context,
      user_params=LLMUserAggregatorParams(
          user_turn_strategies=UserTurnStrategies(
              stop=[
                  TurnAnalyzerUserTurnStopStrategy(
                      turn_analyzer=LocalSmartTurnAnalyzerV3(params=SmartTurnParams())
                  )
              ],
          )
      ),
  )
  ```

  In order to use the user turn strategies you must update to the new
  universal `LLMContext` and `LLMContextAggregatorPair`.
  (PR [#3045](https://github.com/pipecat-ai/pipecat/pull/3045))

- Added `RNNoiseFilter` for real-time noise suppression using RNNoise neural
  network via pyrnnoise library.
  (PR [#3205](https://github.com/pipecat-ai/pipecat/pull/3205))

- Added `GrokRealtimeLLMService` for xAI's Grok Voice Agent API with real-time
  voice conversations:

  - Support for real-time audio streaming with WebSocket connection
  - Built-in server-side VAD (Voice Activity Detection)
  - Multiple voice options: Ara, Rex, Sal, Eve, Leo
  - Built-in tools support: web_search, x_search, file_search
  - Custom function calling with standard Pipecat tools schema
  - Configurable audio formats (PCM at 8kHz-48kHz)
    (PR [#3267](https://github.com/pipecat-ai/pipecat/pull/3267))

- Added an approximation of TTFB for Ultravox.
  (PR [#3268](https://github.com/pipecat-ai/pipecat/pull/3268))

- Added a new `AudioContextTTSService` to the TTS service base classes. The
  `AudioContextWordTTSService` now inherits from `AudioContextTTSService` and
  `WebsocketWordTTSService`.
  (PR [#3289](https://github.com/pipecat-ai/pipecat/pull/3289))

- `LLMUserAggregator` now exposes the following events:

  - `on_user_turn_started`: triggered when a user turn starts
  - `on_user_turn_stopped`: triggered when a user turn ends
  - `on_user_turn_stop_timeout`: triggered when a user turn does not stop
    and times out
    (PR [#3291](https://github.com/pipecat-ai/pipecat/pull/3291))

- Introducing user mute strategies. User mute strategies indicate when user
  input should be muted based on the current system state.

  In conversational agents, user mute strategies are used to prevent user
  input from interrupting bot speech, tool execution, or other critical system
  operations.

  A list of strategies can be specified; all strategies are evaluated for
  every frame so that each strategy can maintain its internal state. A user
  frame is muted if any of the configured strategies indicates it should be
  muted.

  Available user mute strategies:

  - `FirstSpeechUserMuteStrategy`
  - `MuteUntilFirstBotCompleteUserMuteStrategy`
  - `AlwaysUserMuteStrategy`
  - `FunctionCallUserMuteStrategy`

  User mute strategies replace the legacy `STTMuteFilter` and provide a more
  flexible and composable approach to muting user input.

  User mute strategies are configured when setting up the
  `LLMContextAggregatorPair`. For example:

  ```python
  context_aggregator = LLMContextAggregatorPair(
      context,
      user_params=LLMUserAggregatorParams(
          user_mute_strategies=[
              FirstSpeechUserMuteStrategy(),
          ]
      ),
  )
  ```

  In order to use user mute strategies you should update to the new universal
  `LLMContext` and `LLMContextAggregatorPair`.
  (PR [#3292](https://github.com/pipecat-ai/pipecat/pull/3292))

- Added `use_ssl` parameter to `NvidiaSTTService`, `NvidiaSegmentedSTTService`
  and `NvidiaTTSService`.
  (PR [#3300](https://github.com/pipecat-ai/pipecat/pull/3300))

- Added `enable_interruptions` constructor argument to all user turn
  strategies. This tells the `LLMUserAggregator` to push or not push an
  `InterruptionFrame`.
  (PR [#3316](https://github.com/pipecat-ai/pipecat/pull/3316))

- Added `split_sentences` parameter to `SpeechmaticsSTTService` to control
  sentence splitting behavior for finals on sentence boundaries.
  (PR [#3328](https://github.com/pipecat-ai/pipecat/pull/3328))

- Added word-level timestamp support to `AzureTTSService` for accurate
  text-to-audio synchronization.
  (PR [#3334](https://github.com/pipecat-ai/pipecat/pull/3334))

- Added `pronunciation_dict_id` parameter to `CartesiaTTSService.InputParams`
  and `CartesiaHttpTTSService.InputParams` to support Cartesia's pronunciation
  dictionary feature for custom pronunciations.
  (PR [#3346](https://github.com/pipecat-ai/pipecat/pull/3346))

- Added support for using the HeyGen LiveAvatar API with the `HeyGenTransport`
  (see https://www.liveavatar.com/).
  (PR [#3357](https://github.com/pipecat-ai/pipecat/pull/3357))

- Added image support to `OpenAIRealtimeLLMService` via `InputImageRawFrame`:

  - New `start_video_paused` parameter to control initial video input state
  - New `video_frame_detail` parameter to set image processing quality
    ("auto",
    "low", or "high"). This corresponds to OpenAI Realtime's `image_detail`
    parameter.
  - `set_video_input_paused()` method to pause/resume video input at runtime
  - `set_video_frame_detail()` method to adjust video frame quality
    dynamically
  - Automatic rate limiting (1 frame per second) to prevent API overload
    (PR [#3360](https://github.com/pipecat-ai/pipecat/pull/3360))

- Added `UserTurnProcessor`, a frame processor built on `UserTurnController`
  that pushes `UserStartedSpeakingFrame` and `UserStoppedSpeakingFrame` frames
  and interruptions based on the controller's user turn strategies.
  (PR [#3372](https://github.com/pipecat-ai/pipecat/pull/3372))

- Added `UserTurnController` to manage user turns. It emits
  `on_user_turn_started`, `on_user_turn_stopped`, and
  `on_user_turn_stop_timeout` events, and can be integrated into processors to
  detect and handle user turns. `LLMUserAggregator` and `UserTurnProcessor` are
  implemented using this controller.
  (PR [#3372](https://github.com/pipecat-ai/pipecat/pull/3372))

- Added `should_interrupt` property to `DeepgramFluxSTTService`,
  `DeepgramSTTService`, and `SpeechmaticsSTTService` to configure whether the
  bot should be interrupted when the external service detects user speech.
  (PR [#3374](https://github.com/pipecat-ai/pipecat/pull/3374))

- `LLMAssistantAggregator` now exposes the following events:

  - `on_assistant_turn_started`: triggered when the assistant turn starts
  - `on_assistant_turn_stopped`: triggered when the assistant turn ends
  - `on_assistant_thought`: triggered when there's an assistant thought
    available
    (PR [#3385](https://github.com/pipecat-ai/pipecat/pull/3385))

- Added `KrispVivaTurn` analyzer for end of turn detection using the Krisp VIVA
  SDK (requires `krisp_audio`).
  (PR [#3391](https://github.com/pipecat-ai/pipecat/pull/3391))

- Added support for setting up a pipeline task from external files. You can now
  register custom pipeline task setup files by setting the
  `PIPECAT_SETUP_FILES` environment variable. This variable should contain a
  colon-separated list of Python files (e.g. `export
PIPECAT_SETUP_FILES="setup1.py:setup.py:..."`). Each file must define a
  function with the following signature:

  ```python
  async def setup_pipeline_task(task: PipelineTask): ...
  ```

  (PR [#3397](https://github.com/pipecat-ai/pipecat/pull/3397))

- Added a keepalive task for `InworldTTSService` to keep the service connected
  in the event of no generations for longer periods of time.
  (PR [#3403](https://github.com/pipecat-ai/pipecat/pull/3403))

- Added `enable_vad` to `Params` for use in the `GladiaSTTService`. When
  enabled, `GladiaSTTService` acts as the turn controller, emitting
  `UserStartedSpeakingFrame`, `UserStoppedSpeakingFrame`, and optionally
  `InterruptionFrame`.
  (PR [#3404](https://github.com/pipecat-ai/pipecat/pull/3404))

- Added `should_interrupt` property to `GladiaSTTService` to configure whether
  the bot should be interrupted when the external service detects user speech.
  (PR [#3404](https://github.com/pipecat-ai/pipecat/pull/3404))

- Added `VonageFrameSerializer` for the Vonage Video API Audio Connector
  WebSocket protocol.
  (PR [#3410](https://github.com/pipecat-ai/pipecat/pull/3410))

- Added `append_trailing_space` parameter to `TTSService` to automatically
  append a trailing space to text before sending to TTS, helping prevent some
  services from vocalizing trailing punctuation.
  (PR [#3424](https://github.com/pipecat-ai/pipecat/pull/3424))

### Changed

- Updated `ElevenLabsRealtimeSTTService` to accept the
  `include_language_detection` parameter to detect language.

  ```python
  stt = ElevenLabsRealtimeSTTService(
      api_key=os.getenv("ELEVENLABS_API_KEY"), include_language_detection=True
  )
  ```

  (PR [#3216](https://github.com/pipecat-ai/pipecat/pull/3216))

- Updated `SpeechmaticsSTTService` to use new Python Voice SDK with improved
  VAD, Smart Turn capabilities, and brings dramatic improvements to latency
  without any impact on accuracy. Use the `turn_detection_mode` parameter to control
  the endpointing of speech, with `TurnDetectionMode.EXTERNAL` (default),
  `TurnDetectionMode.ADAPTIVE`, or `TurnDetectionMode.SMART_TURN`.

  ```python
      stt = SpeechmaticsSTTService(
          api_key=os.getenv("SPEECHMATICS_API_KEY"),
          params=SpeechmaticsSTTService.InputParams(
              language=Language.EN,
              turn_detection_mode=SpeechmaticsSTTService.TurnDetectionMode.ADAPTIVE,
              speaker_active_format="<{speaker_id}>{text}</{speaker_id}>",
          ),
      )
  ```

  (PR [#3225](https://github.com/pipecat-ai/pipecat/pull/3225))

- `daily-python` updated to 0.23.0.
  (PR [#3257](https://github.com/pipecat-ai/pipecat/pull/3257))

- `TranscriptionFrame` and `InterimTranscriptionFrame` produced by
  `DailyTransport` now include the transport source (i.e., the originating
  audio track).
  (PR [#3257](https://github.com/pipecat-ai/pipecat/pull/3257))

- Updates to Inworld TTS services:

  - Improved `InworldTTSService`'s websocket implementation to better flush
    and close context to better handle long inputs.
  - Improved docstrings for `InworldTTSService` and `InworldHttpTTSService`.
    (PR [#3288](https://github.com/pipecat-ai/pipecat/pull/3288))

- Improved the error handling and reconnection logic for `WebsocketServer` by
  distinguishing between errors when disconnecting and websocket communication
  errors.
  (PR [#3392](https://github.com/pipecat-ai/pipecat/pull/3392))

- Updated `DeepgramSTTService` to push user started/stopped speaking and
  interruption frames when `vad_enabled` is set to true. This centralizes the
  frames into the service, removing the need to have your application code
  handle Deepgram's events and push these frames.
  (PR [#3314](https://github.com/pipecat-ai/pipecat/pull/3314))

- Added encoding validation to `DeepgramTTSService` to prevent unsupported
  encodings from reaching the API. The service now raises `ValueError` at
  initialization with a clear error message.
  (PR [#3329](https://github.com/pipecat-ai/pipecat/pull/3329))

- Updated `read_audio_frame` & `read_video_frame` methods in
  `SmallWebRTCClient` to check if the track is enabled before logging a
  warning.
  (PR [#3336](https://github.com/pipecat-ai/pipecat/pull/3336))

- Updated `CartesiaTTSService` to support setting `language=None`, resulting in
  Cartesia auto-detecting the language of the conversation.
  (PR [#3366](https://github.com/pipecat-ai/pipecat/pull/3366))

- The bundled Smart Turn weights are now updated to v3.2, which has better
  handling of short utterances, and is more robust against background noise.
  (PR [#3367](https://github.com/pipecat-ai/pipecat/pull/3367))

- Updated `SpeechmaticsSTTService` dependency to `speechmatics-voice[smart]>=0.2.6`
  (PR [#3371](https://github.com/pipecat-ai/pipecat/pull/3371))

- Smart Turn now takes into account `vad_start_seconds` when buffering audio,
  meaning that the start of the turn audio is not cut off. This improves
  accuracy for short utterances.

- The default value of `pre_speech_ms` is now set to 500ms for Smart Turn.
  (PR [#3377](https://github.com/pipecat-ai/pipecat/pull/3377))

- Improved Krisp SDK management to allow `KrispVivaTurn` and `KrispVivaFilter`
  to share a single SDK instance within the same process.
  (PR [#3391](https://github.com/pipecat-ai/pipecat/pull/3391))

- Updated default model for `GroqTTSService` to `canopylabs/orpheus-v1-english`
  and voice ID to `autumn`.
  (PR [#3399](https://github.com/pipecat-ai/pipecat/pull/3399))

- Enhanced `FastAPIWebsocketTransport` with optional protocol-level audio
  packetization via the `fixed_audio_packet_size` parameter to support media
  endpoints requiring strict framing and real-time pacing.
  (PR [#3410](https://github.com/pipecat-ai/pipecat/pull/3410))

- `DeepgramTTSService` and `RimeTTSService` now set `append_trailing_space` to
  `True` to prevent punctuation (e.g., “dot”) from being pronounced.
  (PR [#3424](https://github.com/pipecat-ai/pipecat/pull/3424))

- Updated `GeminiLiveLLMService` to push `LLMThoughtStartFrame`,
  `LLMThoughtTextFrame`, and `LLMThoughtEndFrame` when the model returns
  thought content.
  (PR [#3431](https://github.com/pipecat-ai/pipecat/pull/3431))

### Deprecated

- `pipecat.audio.interruptions.MinWordsInterruptionStrategy` is deprecated. Use
  `pipecat.turns.user_start.MinWordsUserTurnStartStrategy` with
  `LLMUserAggregator`'s new `user_turn_strategies` parameter instead.
  (PR [#3045](https://github.com/pipecat-ai/pipecat/pull/3045))

- `FrameProcessor.interruption_strategies` is deprecated, use
  `LLMUserAggregator`'s new `user_turn_strategies` parameter instead.
  (PR [#3045](https://github.com/pipecat-ai/pipecat/pull/3045))

- The `LLMUserAggregatorParams` and `LLMAssistantAggregatorParams` classes in
  `pipecat.processors.aggregators.llm_response` are now deprecated. Use the new
  universal `LLMContext` and `LLMContextAggregatorPair` instead.
  (PR [#3045](https://github.com/pipecat-ai/pipecat/pull/3045))

- Deprecated the `emulated` field in the `UserStartedSpeakingFrame` and
  `UserStoppedSpeakingFrame` frames.
  (PR [#3045](https://github.com/pipecat-ai/pipecat/pull/3045))

- `EmulateUserStartedSpeakingFrame` and `EmulateUserStoppedSpeakingFrame`
  frames are deprecated.
  (PR [#3045](https://github.com/pipecat-ai/pipecat/pull/3045))

- ⚠️ `TransportParams.turn_analyzer` is deprecated and might result in
  unexpected behavior, use `LLMUserAggregator`'s new `user_turn_strategies`
  parameter instead.
  (PR [#3045](https://github.com/pipecat-ai/pipecat/pull/3045))

- For `SpeechmaticsSTTService`, the `end_of_utterance_mode` parameter is
  deprecated. Use the new `turn_detection_mode` parameter instead, with
  `TurnDetectionMode.EXTERNAL`,`TurnDetectionMode.ADAPTIVE`, or
  `TurnDetectionMode.SMART_TURN`. The `enable_vad` parameter is also
  deprecated and is inferred from the `turn_detection_mode`.
  (PR [#3225](https://github.com/pipecat-ai/pipecat/pull/3225))

- `OpenAILLMContext` and its associated things (context aggregators, etc.) are
  now deprecated in favor of the universal `LLMContext` and its associated
  things.

  From the developer's point of view, switching to using `LLMContext`
  machinery will usually be a matter of going from this:

  ```python
  context = OpenAILLMContext(messages, tools)
  context_aggregator = llm.create_context_aggregator(context)
  ```

  To this:

  ```
  context = LLMContext(messages, tools)
  context_aggregator = LLMContextAggregatorPair(context)
  ```

  (PR [#3263](https://github.com/pipecat-ai/pipecat/pull/3263))

- `STTMuteFilter` is deprecated and will be removed in a future version. Use
  `LLMUserAggregator`'s new `user_mute_strategies` instead.
  (PR [#3292](https://github.com/pipecat-ai/pipecat/pull/3292))

- `FrameProcessor.interruptions_allowed` is now deprecated, use
  `LLMUserAggregator`'s new parameter `user_mute_strategies` instead.
  (PR [#3297](https://github.com/pipecat-ai/pipecat/pull/3297))

- `PipelineParams.allow_interruptions` is now deprecated, use
  `LLMUserAggregator`'s new parameter `user_turn_strategies` instead. For
  example, to disable interruptions but still get user turns you can do:

  ```python
  context_aggregator = LLMContextAggregatorPair(
      context,
      user_params=LLMUserAggregatorParams(
          user_turn_strategies=UserTurnStrategies(
              start=[TranscriptionUserTurnStartStrategy(enable_interruptions=False)],
          ),
      ),
  )
  ```

  (PR [#3297](https://github.com/pipecat-ai/pipecat/pull/3297))

- `TranscriptProcessor` and related data classes and frames
  (`TranscriptionMessage`, `ThoughtTranscriptionMessage`,
  `TranscriptionUpdateFrame`) are deprecated. Use `LLMUserAggregator`'s and
  `LLMAssistantAggregator`'s new events (`on_user_turn_stopped` and
  `on_assistant_turn_stopped`) instead.
  (PR [#3385](https://github.com/pipecat-ai/pipecat/pull/3385))

- Deprecated support for the `vad_events` `LiveOptions` in
  `DeepgramSTTService`. Instead, use a local Silero VAD for VAD events.
  Additionally, deprecated `should_interrupt` which will be removed along with
  `vad_events` support in a future release.
  (PR [#3386](https://github.com/pipecat-ai/pipecat/pull/3386))

- Loading external observers from files is deprecated, use the new pipeline
  task setup files and `PIPECAT_SETUP_FILES` environment variable instead.
  (PR [#3397](https://github.com/pipecat-ai/pipecat/pull/3397))

### Fixed

- Improved error handling in `ElevenLabsRealtimeSTTService`
  (PR [#3233](https://github.com/pipecat-ai/pipecat/pull/3233))

- Fixed an issue in `ElevenLabsRealtimeSTTService` causing an infinite loop
  that blocks the process if the websocket disconnects due to an error
  (PR [#3233](https://github.com/pipecat-ai/pipecat/pull/3233))

- Fixed a bug in `STTMuteFilter` where the user was not always muted during
  function calls, especially when there were multiple simultaneous calls.
  (PR [#3292](https://github.com/pipecat-ai/pipecat/pull/3292))

- Fixed a `RNNoiseFilter` issue that would cause a "[Errno 12] Cannot allocate
  memory" error when processing silence audio frames.
  (PR [#3322](https://github.com/pipecat-ai/pipecat/pull/3322))

- Updated `SpeechmaticsSTTService` for version `0.0.99+`:

  - Fixed `SpeechmaticsSTTService` to listen for `VADUserStoppedSpeakingFrame`
    in order to finalize transcription.
  - Default to `TurnDetectionMode.FIXED` for Pipecat-controlled end of turn
    detection.
  - Only emit VAD + interruption frames if VAD is enabled within the plugin
    (modes other than `TurnDetectionMode.FIXED` or `TurnDetectionMode.EXTERNAL`).
    (PR [#3328](https://github.com/pipecat-ai/pipecat/pull/3328))

- Fixed an issue with function calling where a handler failing to invoke its
  result callback could leave the context stuck in IN_PROGRESS, causing LLM
  inference for subsequent function call results to block while waiting on the
  unresolved call.
  (PR [#3343](https://github.com/pipecat-ai/pipecat/pull/3343))

- Fixed an issue with DeepgramTTSService where the model would output "Dot"
  instead of a period in some circumstances.
  (PR [#3345](https://github.com/pipecat-ai/pipecat/pull/3345))

- Fixed an issue in `traced_stt` where `model_name` in OpenTelemetry appears as
  `unknown`.
  (PR [#3351](https://github.com/pipecat-ai/pipecat/pull/3351))

- Fixed an issue in GeminiLiveLLMService where TranscriptionFrames were
  occasionally not pushed.
  (PR [#3356](https://github.com/pipecat-ai/pipecat/pull/3356))

- Fixed potential memory leaks and initialization issues in `KrispVivaFilter`
  by improving SDK lifecycle management.
  (PR [#3391](https://github.com/pipecat-ai/pipecat/pull/3391))

- Fixed timing issue in `BaseOutputTransport` where the bot speaking flag was
  set after awaiting, allowing the event loop to re-enter the method before the
  guard was set.
  (PR [#3400](https://github.com/pipecat-ai/pipecat/pull/3400))

- Fixed parallel function calling when using Gemini thinking.
  (PR [3420](https://github.com/pipecat-ai/pipecat/pull/3420))

- Fixed an issue in `traced_llm` where `model_name` in OpenTelemetry appears as
  `unknown`.
  (PR [#3422](https://github.com/pipecat-ai/pipecat/pull/3422))

- Fixed an issue in `traced_tts`, `traced_gemini_live`, and
  `traced_openai_realtime` where `model_name` in OpenTelemetry appears as
  `unknown`.
  (PR [#3428](https://github.com/pipecat-ai/pipecat/pull/3428))

- Fixed `request_image_frame` (for backwards compatibility) and restored
  function-call–related fields in `UserImageRequestFrame` and
  `UserImageRawFrame`, preventing a case where adding a non-LLM message to the
  context could trigger duplicate LLM inferences (on image arrival and on
  function-call result), potentially causing an infinite inference loop.
  (PR [#3430](https://github.com/pipecat-ai/pipecat/pull/3430))

- Fixed `LLMContext.create_audio_message()` by correcting an internal helper
  that was incorrectly declared async while being run in `asyncio.to_thread()`.
  (PR [#3435](https://github.com/pipecat-ai/pipecat/pull/3435))

### Other

- Added `52-live-transcription.py` foundational example demonstrating live
  transcription and translation from English to Spanish. In this example, the
  bot is not interruptible: as the user continues speaking, English
  transcriptions are queued, and the bot continuously translates and speaks
  each queued sentence in Spanish without being interrupted by new user speech.
  (PR [#3316](https://github.com/pipecat-ai/pipecat/pull/3316))

- Added a new foundational example `53-concurrent-llm-evaluation.py` that shows
  how to use `UserTurnProcessor`.
  (PR [#3372](https://github.com/pipecat-ai/pipecat/pull/3372))

- Added a new foundational example `28-user-assistant-turns.py` that shows how
  to use the new `LLMUserAggregator` and `LLMAssistantAggregator` events to
  gather a conversation transcript.
  (PR [#3385](https://github.com/pipecat-ai/pipecat/pull/3385))

## [0.0.98] - 2025-12-17

### Added

- Added `RimeNonJsonTTSService` which supports non-JSON streaming mode. This
  new class supports websocket streaming for the Arcana model.
  (PR [#3085](https://github.com/pipecat-ai/pipecat/pull/3085))

- Added additional functionality related to "thinking", for Google and
  Anthropic LLMs.

  1. New typed parameters for Google and Anthropic LLMs that control the
     models' thinking behavior (like how much thinking to do, and whether to
     output thoughts or thought summaries):
     - `AnthropicLLMService.ThinkingConfig`
     - `GoogleLLMService.ThinkingConfig`
  2. New frames for representing thoughts output by LLMs:
     - `LLMThoughtStartFrame`
     - `LLMThoughtTextFrame`
     - `LLMThoughtEndFrame`
  3. A generic mechanism for recording LLM thoughts to context, used
     specifically to support Anthropic, whose thought signatures are expected
     to appear alongside the text of the thoughts within assistant context
     messages. See:
     - `LLMThoughtEndFrame.signature`
     - `LLMAssistantAggregator` handling of the above field
     - `AnthropicLLMAdapter` handling of `"thought"` context messages
  4. Google-specific logic for inserting thought signatures into the context,
     to help maintain thinking continuity in a chain of LLM calls. See:
     - `GoogleLLMService` sending `LLMMessagesAppendFrame`s to add
       LLM-specific
       `"thought_signature"` messages to context
     - `GeminiLLMAdapter` handling of `"thought_signature"` messages
  5. An expansion of `TranscriptProcessor` to process LLM thoughts in
     addition to user and assistant utterances. See:
     - `TranscriptProcessor(process_thoughts=True)` (defaults to `False`)
     - `ThoughtTranscriptionMessage`, which is now also emitted with the
       `"on_transcript_update"` event
       (PR [#3175](https://github.com/pipecat-ai/pipecat/pull/3175))

- Data and control frames can now be marked as non-interruptible by using the
  `UninterruptibleFrame` mixin. Frames marked as `UninterruptibleFrame` will
  not be interrupted during processing, and any queued frames of this type will
  be retained in the internal queues. This is useful when you need ordered
  frames (data or control) that should not be discarded or cancelled due to
  interruptions.
  (PR [#3189](https://github.com/pipecat-ai/pipecat/pull/3189))

- Added `on_conversation_detected` event to `VoicemaiDetector`.
  (PR [#3207](https://github.com/pipecat-ai/pipecat/pull/3207))

- Added `x-goog-api-client` header with Pipecat's version to all Google
  services' requests.
  (PR [#3208](https://github.com/pipecat-ai/pipecat/pull/3208))

- Added support for the HeyGen LiveAvatar API (see https://www.liveavatar.com/).
  (PR [#3210](https://github.com/pipecat-ai/pipecat/pull/3210))

- Added to `AWSNovaSonicLLMService` functionality related to the new (and now
  default) Nova 2 Sonic model (`"amazon.nova-2-sonic-v1:0"`):

  - Added the `endpointing_sensitivity` parameter to control how quickly the
    model decides the user has stopped speaking.
  - Made the assistant-response-trigger hack a no-op. It's only needed for
    the older Nova Sonic model.
    (PR [#3212](https://github.com/pipecat-ai/pipecat/pull/3212))

- [Ultravox Realtime](https://docs.ultravox.ai) is now a supported
  speech-to-speech service.

  - Added `UltravoxRealtimeLLMService` for the integration.
  - Added `49-ultravox-realtime.py` example (with tool calling).
    (PR [#3227](https://github.com/pipecat-ai/pipecat/pull/3227))

- Added Daily PSTN dial-in support to the development runner with `--dialin`
  flag. This includes:

  - `/daily-dialin-webhook` endpoint that handles incoming Daily PSTN webhooks
  - Automatic Daily room creation with SIP configuration
  - `DialinSettings` and `DailyDialinRequest` types in `pipecat.runner.types`
    for type-safe dial-in data
  - The runner now mimics Pipecat Cloud's dial-in webhook handling for local
    development
    (PR [#3235](https://github.com/pipecat-ai/pipecat/pull/3235))

- Add Gladia session id to logs for `GladiaSTTService`.
  (PR [#3236](https://github.com/pipecat-ai/pipecat/pull/3236))

- Added `InworldHttpTTSService` which uses Inworld's HTTP based TTS service in
  either streaming or non-streaming mode. Note: This class was previously named
  `InworldTTSService`.
  (PR [#3239](https://github.com/pipecat-ai/pipecat/pull/3239))

- Added `language_hints_strict` parameter to `SonioxSTTService` to strictly
  enforces language hints. This ensures that transcription occurs in the
  specified language.
  (PR [#3245](https://github.com/pipecat-ai/pipecat/pull/3245))

- Added Pipecat library version info to the `about` field in the `bot-ready`
  RTVI message.
  (PR [#3248](https://github.com/pipecat-ai/pipecat/pull/3248))

- Added `VisionFullResponseStartFrame`, `VisionFullResponseEndFrame` and
  `VisionTextFrame`. This are used by vision services similar to LLM
  services.
  (PR [#3252](https://github.com/pipecat-ai/pipecat/pull/3252))

### Changed

- `FunctionCallInProgressFrame` and `FunctionCallResultFrame` have changed from
  system frames to a control frame and a data frame, respectively, and are
  now both marked as `UninterruptibleFrame`.
  (PR [#3189](https://github.com/pipecat-ai/pipecat/pull/3189))

- `UserBotLatencyLogObserver` now uses `VADUserStartedSpeakingFrame` and
  `VADUserStoppedSpeakingFrame` to determine latency from user stopped speaking
  to bot started speaking.
  (PR [#3206](https://github.com/pipecat-ai/pipecat/pull/3206))

- Updated `HeyGenVideoService` and `HeyGenTransport` to support both HeyGen
  APIs (Interactive Avatar and Live Avatar).
  Using them is as simple as specifying the `service_type` when creating the
  `HeyGenVideoService` and the `HeyGenTransport`:

  ```python
  heyGen = HeyGenVideoService(
      api_key=os.getenv("HEYGEN_LIVE_AVATAR_API_KEY"),
      service_type=ServiceType.LIVE_AVATAR,
      session=session,
  )
  ```

  (PR [#3210](https://github.com/pipecat-ai/pipecat/pull/3210))

- Made `"amazon.nova-2-sonic-v1:0"` the new default model for
  `AWSNovaSonicLLMService`.
  (PR [#3212](https://github.com/pipecat-ai/pipecat/pull/3212))

- Updated the `run_inference` methods in the LLM service classes
  (`AnthropicLLMService`, `AWSBedrockLLMService`, `GoogleLLMService`, and
  `OpenAILLMService` and its base classes) to use the provided LLM
  configuration parameters.
  (PR [#3214](https://github.com/pipecat-ai/pipecat/pull/3214))

- Updated default models for:

  - `GeminiLiveLLMService` to `gemini-2.5-flash-native-audio-preview-12-2025`.
  - `GeminiLiveVertexLLMService` to `gemini-live-2.5-flash-native-audio`.
    (PR [#3228](https://github.com/pipecat-ai/pipecat/pull/3228))

- Changed the `reason` field in `EndFrame`, `CancelFrame`, `EndTaskFrame`, and
  `CancelTaskFrame` from `str` to `Any` to indicate that it can hold values
  other than strings.
  (PR [#3231](https://github.com/pipecat-ai/pipecat/pull/3231))

- Updated websocket STT services to use the `WebsocketSTTService` base class.
  This base class manages the websocket connection and handles reconnects.
  Updated services:

  - `AssemblyAISTTService`
  - `AWSTranscribeSTTService`
  - `GladiaSTTService`
  - `SonioxSTTService`
    (PR [#3236](https://github.com/pipecat-ai/pipecat/pull/3236))

- Changed Inworld's TTS service implementations:

  - Previously, the HTTP implementation was named `InworldTTSService`. That
    has been moved to `InworldHttpTTSService`. This service now supports
    word-timestamp alignment data in both streaming and non-streaming modes.
  - Updated the `InworldTTSService` class to use Inworld's Websocket API.
    This class now has support for word-timestamp alignment data and tracks
    contexts for each user turn.
    (PR [#3239](https://github.com/pipecat-ai/pipecat/pull/3239))

- ⚠️ Breaking change: `WordTTSService.start_word_timestamps()` and
  `WordTTSService.reset_word_timestamps()` are now async.
  (PR [#3240](https://github.com/pipecat-ai/pipecat/pull/3240))

- Updated the current RTVI version to 1.1.0 to reflect recent additions and
  deprecations.

  - New RTVI Messages: `send-text` and `bot-output`
  - Deprecated Messages: `append-to-context` and `bot-transcription`
    (PR [#3248](https://github.com/pipecat-ai/pipecat/pull/3248))

- `MoondreamService` now pushes `VisionFullResponseStartFrame`,
  `VisionFullResponseEndFrame` and `VisionTextFrame`.
  (PR [#3252](https://github.com/pipecat-ai/pipecat/pull/3252))

### Deprecated

- `FalSmartTurnAnalyzer` and `LocalSmartTurnAnalyzer` are deprecated and will
  be removed in a future version. Use `LocalSmartTurnAnalyzerV3` instead.
  (PR [#3219](https://github.com/pipecat-ai/pipecat/pull/3219))

### Removed

- Removed the deprecated VLLM-based open source Ultravox STT service.
  (PR [#3227](https://github.com/pipecat-ai/pipecat/pull/3227))

### Fixed

- Fixed a bug in `AWSNovaSonicLLMService` where we would mishandle cancelled
  tool calls in the context, resulting in errors.
  (PR [#3212](https://github.com/pipecat-ai/pipecat/pull/3212))

- Better support conversation history with Gemini 2.5 Flash Image (model
  "gemini-2.5-flash-image"). Prior to this fix, the model had no memory of
  previous images it had generated, so it wouldn't be able to iterate on
  them.
  (PR [#3224](https://github.com/pipecat-ai/pipecat/pull/3224))

- Support conversations with Gemini 3 Pro Image (model
  "gemini-3-pro-image-preview"). Prior to this fix, after the model generated
  an image the conversation would not be able to progress.
  (PR [#3224](https://github.com/pipecat-ai/pipecat/pull/3224))

- Fixed an issue where `ElevenLabsHttpTTSService` was not updating
  voice settings when receiving a `TTSUpdateSettingsFrame`.
  (PR [#3226](https://github.com/pipecat-ai/pipecat/pull/3226))

- Fixed the return type for `SmallWebRTCRequestHandler.handle_web_request()`
  function.
  (PR [#3230](https://github.com/pipecat-ai/pipecat/pull/3230))

- Fix a bug in LLM context audio content handling
  (PR [#3234](https://github.com/pipecat-ai/pipecat/pull/3234))

- In `GladiaSTTService`, reset the `_bytes_sent` counter on connecting the
  websocket. This avoids unnecessary audio buffer trimming.
  (PR [#3236](https://github.com/pipecat-ai/pipecat/pull/3236))

- Fixed a TTS service word-timestamp issue that could cause generated
  `TTSTextFrame` instances to have an incorrect pts (`pts = -1`).
  (PR [#3240](https://github.com/pipecat-ai/pipecat/pull/3240))

- Fixed an issue in `SimpleTextAggreagtor` where spaces were not being stripped
  before returning the aggregation. This resulted in an extra space for TTS
  services that don't support word-timestamp alignment data.
  (PR [#3247](https://github.com/pipecat-ai/pipecat/pull/3247))

## [0.0.97] - 2025-12-05

### Added

- Added new Gradium services, `GradiumSTTService` and `GradiumTTSService`, for
  speech-to-text and text-to-speech functionality using Gradium's API.

- Additions for `AsyncAITTSService` and `AsyncAIHttpTTSService`:

  - Added new `languages`: `pt`, `nl`, `ar`, `ru`, `ro`, `ja`, `he`, `hy`,
    `tr`, `hi`, `zh`.
  - Updated the default model to `asyncflow_multilingual_v1.0` for improved
    accuracy and broader language coverage.

- Added optional tool and tool output filters for MCP services.

### Changed

- Updated Deepgram logging to include Deepgram request IDs for improved
  debugging.

- Text Aggregation Improvements:

  - **Breaking Change**: `BaseTextAggregator.aggregate()` now returns
    `AsyncIterator[Aggregation]` instead of `Optional[Aggregation]`. This
    enables the aggregator to return multiple results based on the provided
    text.
  - Refactored text aggregators to use inheritance: `SkipTagsAggregator` and
    `PatternPairAggregator` now inherit from `SimpleTextAggregator`, reusing
    the base class's sentence detection logic.

- Improved interruption handling to prevent bots from repeating themselves. LLM
  services that return multiple sentences in a single response (e.g.,
  `GoogleLLMService`) are now split into individual sentences before being sent
  to TTS. This ensures interruptions occur at sentence boundaries, preventing
  the bot from repeating content after being interrupted during long responses.

- Updated `AICFilter` to use Quail STT as the default model
  (`AICModelType.QUAIL_STT`). Quail STT is optimized for human-to-machine
  interaction (e.g., voice agents, speech-to-text) and operates at a native
  sample rate of 16 kHz with fixed enhancement parameters.

- If an unexpected exception is caught, or if `FrameProcessor.push_error()` is
  called with an exception, the file name and line number where the exception
  occured are now logged.

- Updated Smart Turn model weights to v3.1.

- Smart Turn analyzer now uses the full context of the turn rather than just
  the audio since VAD last triggered.

- Updated `CartesiaSTTService` to return the full transcription `result` in the
  `TranscriptionFrame` and `InterimTranscriptionFrame`. This provides access to
  word timestamp data.

- `HumeTTSService` changes:

  - Added tracking headers (`X-Hume-Client-Name` and `X-Hume-Client-Version`)
    to all requests made by `HumeTTSService` to the Hume API for better usage
    tracking and analytics.
  - Added `stop()` and `cancel()` cleanup methods to `HumeTTSService` to
    properly close the HTTP client and prevent resource leaks.

### Deprecated

- NVIDIA Services name changes (all functionality is unchanged):

  - `NimLLMService` is now deprecated, use `NvidiaLLMService` instead.
  - `RivaSTTService` is now deprecated, use `NvidiaSTTService` instead.
  - `RivaTTSService` is now deprecated, use `NvidiaTTSService` instead.
  - Use `uv pip install pipecat-ai[nvidia]` instead of
    `uv pip install pipecat-ai[riva]`

- The `noise_gate_enable` parameter in `AICFilter` is deprecated and no longer
  has any effect. Noise gating is now handled automatically by the AIC VAD
  system. Use `AICFilter.create_vad_analyzer()` for VAD functionality instead.

- Package `pipecat.sync` is deprecated, use `pipecat.utils.sync` instead.

### Fixed

- Fixed bug in `PatternPairAggregator` where pattern handlers could be called
  multiple times for `KEEP` or `AGGREGATE` patterns.

- Fixed sentence aggregation to correctly handle ambiguous punctuation in
  streaming text, such as currency ("$29.95") and abbreviations ("Mr. Smith").

- Fixed an issue in `AWSTranscribeSTTService` where the `region` arg was always
  set to `us-east-1` when providing an AWS_REGION env var.

- Fixed an issue in `SarvamTTSService` where the last sentence was not being
  spoken. Now, audio is flushed when the TTS services receives the
  `LLMFullResponseEndFrame` or `EndFrame`.

- Fixed an issue in `DeepgramTTSService` where a `TTSStoppedFrame` was
  incorrectly pushed after a functional call. This caused an issue with the
  voice-ui-kit's conversational panel rending of the LLM output after a
  function call.

- Fixed an issue where `LLMTextFrame.skip_tts` was being overwritten by LLM
  services.

- Fixed an issue that caused `WebsocketService` instances to attempt
  reconnection during shutdown.

- Fixed an issue in `ElevenLabsTTSService` where character usage metrics were
  only reported on the first TTS generation per turn.

## [0.0.96] - 2025-11-26 🦃 "Happy Thanksgiving!" 🦃

### Added

- Added `AWSBedrockAgentCoreProcessor` to support invoking an AgentCore-hosted
  agent in a Pipecat pipeline.

- Enhanced error handling across the framework:

  - Added `on_error` callback to `FrameProcessor` for centralized error
    handling.

  - Renamed `push_error(error: ErrorFrame)` to `push_error_frame(error: ErrorFrame)`
    for clarity.

  - Added new `push_error` method for simplified error reporting:

    ```python
    async def push_error(error_msg: str,
                         exception: Optional[Exception] = None,
                         fatal: bool = False)
    ```

  - Standardized error logging by replacing `logger.exception` calls with
    `logger.error` throughout the codebase.

- Added `cache_read_input_tokens`, `cache_creation_input_tokens` and
  `reasoning_tokens` to OTel spans for LLM call

- Added `LiveKitRESTHelper` utility class for managing LiveKit rooms via REST API.

- Added `DeepgramSageMakerSTTService` which connects to a SageMaker hosted
  Deepgram STT model. Added `07c-interruptible-deepgram-sagemaker.py`
  foundational example.

- Added `SageMakerBidiClient` to connect to SageMaker hosted BiDi compatible
  services.

- Added support for `include_timestamps` and `enable_logging` in
  `ElevenLabsRealtimeSTTService`. When `include_timestamps` is enabled,
  timestamp data is included in the `TranscriptionFrame`'s `result`
  parameter.

- Added optional speaking rate control to `InworldTTSService`.

- Introduced a new `AggregatedTextFrame` type to support passing text along with
  an `aggregated_by` field to describe the type of text
  included. `TTSTextFrame`s now inherit from `AggregatedTextFrame`. With this
  inheritance, an observer can watch for `AggregatedTextFrame`s to accumlate the
  perceived output and determine whether or not the text was spoken based on if
  that frame is also a `TTSTextFrame`.

  With this frame, the llm token stream can be transformed into custom
  composable chunks, allowing for aggregation outside the TTS service. This
  makes it possible to listen for or handle those aggregations and sets the
  stage for doing things like composing a best effort of the perceived llm
  output in a more digestable form and to do so whether or not it is processed
  by a TTS or if even a TTS exists.

- Introduced `LLMTextProcessor`: A new processor meant to allow customization
  for how LLMTextFrames should be aggregated and considered. It's purpose is to
  turn `LLMTextFrame`s into `AggregatedTextFrame`s. By default, a TTSService
  will still aggregate `LLMTextFrame`s by sentence for the service to
  consume. However, if you wish to override how the llm text is aggregated, you
  should no longer override the TTS's internal text_aggregator, but instead,
  insert this processor between your LLM and TTS in the pipeline.

- New `bot-output` RTVI message to represent what the bot actually "says".

  - The `RTVIObserver` now emits `bot-output` messages based off the new
    `AggregatedTextFrame`s (`bot-tts-text` and `bot-llm-text` are still
    supported and generated, but `bot-transcript` is now deprecated in lieu of
    this new, more thorough, message).

  - The new `RTVIBotOutputMessage` includes the fields:

    - `spoken`: A boolean indicating whether the text was spoken by TTS

    - `aggregated_by`: A string representing how the text was aggregated
      ("sentence", "word", "my custom aggregation")

  - Introduced new fields to `RTVIObserver` to support the new `bot-output`
    messaging:

    - `bot_output_enabled`: Defaults to True. Set to false to disable bot-output
      messages.

    - `skip_aggregator_types`: Defaults to `None`. Set to a list of strings that
      match aggregation types that should not be included in bot-output
      messages. (Ex. `credit_card`)

  - Introduced new methods, `add_text_transformer()` and
    `remove_text_transformer()`, to `RTVIObserver` to support providing (and
    subsequently removing) callbacks for various types of aggregations (or all
    aggregations with `*`) that can modify the text before being sent as a
    `bot-output` or `tts-text` message. (Think obscuring the credit card or
    inserting extra detail the client might want that the context doesn't need.)

- In `MiniMaxHttpTTSService`:

  - Added support for speech-2.6-hd and speech-2.6-turbo models

  - Added languages: Afrikaans, Bulgarian, Catalan, Danish, Persian, Filipino,
    Hebrew, Croatian, Hungarian, Malay, Norwegian, Nynorsk, Slovak, Slovenian,
    Swedish, and Tamil

  - Added new emotions: calm and fluent

- Added `enable_logging` to `SimliVideoService` input parameters. It's disabled
  by default.

### Changed

- Updated `FishAudioTTSService` default model to `s1`.

- Updated `DeepgramTTSService` to use Deepgram's TTS websocket API. ⚠️ This is
  a potential breaking change, which only affects you if you're self-hosting
  `DeepgramTTSService`. The new service uses Websockets and improves TTFB
  latency.

- Updated `daily-python` to 0.22.0.

- `BaseTextAggregator` changes:

  Modified the BaseTextAggregator type so that when text gets aggregated,
  metadata can be associated with it. Currently, that just means a `type`, so
  that the aggregation can be classified or described. Changes made to support
  this:

  - ⚠️ IMPORTANT: Aggregators are now expected to strip leading/trailing white
    space characters before returning their aggregation from `aggregation()` or
    `.text`. This way all aggregators have a consistent contract allowing
    downstream use to know how to stitch aggregations back together.

  - Introduced a new `Aggregation` dataclass to represent both the aggregated
    `text` and a string identifying the `type` of aggregation (ex. "sentence",
    "word", "my custom aggregation")

  - ⚠️ Breaking change: `BaseTextAggregator.text` now returns an `Aggregation`
    (instead of `str`).

    Before:

    ```python
    aggregated_text = myAggregator.text
    ```

    Now:

    ```python
    aggregated_text = myAggregator.text.text
    ```

  - ⚠️ Breaking change: `BaseTextAggregator.aggregate()` now returns
    `Optional[Aggregation]` (instead of `Optional[str]`).

    Before:

    ```python
    aggregation = myAggregator.aggregate(text)
    print(f"successfully aggregated text: {aggregation}")
    ```

    Now:

    ```python
    aggregation = myAggregator.aggregate(text)
    if aggregation:
        print(f"successfully aggregated text: {aggregation.text}")
    ```

  - `SimpleTextAggregator`, `SkipTagsAggregator`, `PatternPairAggregator`
    updated to produce/consume `Aggregation` objects.

  - All uses of the above Aggregators have been updated accordingly.

- Augmented the `PatternPairAggregator` so that matched patterns can be treated
  as their own aggregation, taking advantage of the new. To that end:

  - Introduced a new, preferred version of `add_pattern` to support a new option
    for treating a match as a separate aggregation returned from
    `aggregate()`. This replaces the now deprecated `add_pattern_pair` method
    and you provide a `MatchAction` in lieu of the `remove_match` field.

    - `MatchAction` enum: `REMOVE`, `KEEP`, `AGGREGATE`, allowing customization
      for how a match should be handled.

      - `REMOVE`: The text along with its delimiters will be removed from the
        streaming text. Sentence aggregation will continue on as if this text
        did not exist.

      - `KEEP`: The delimiters will be removed, but the content between them
        will be kept. Sentence aggregation will continue on with the internal
        text included.

      - `AGGREGATE`: The delimiters will be removed and the content between will
        be treated as a separate aggregation. Any text before the start of the
        pattern will be returned early, whether or not a complete sentence was
        found. Then the pattern will be returned. Then the aggregation will
        continue on sentence matching after the closing delimiter is found. The
        content between the delimiters is not aggregated by sentence. It is
        aggregated as one single block of text.

    - `PatternMatch` now extends `Aggregation` and provides richer info to
      handlers.

  - ⚠️ Breaking change: The `PatternMatch` type returned to handlers registered
    via `on_pattern_match` has been updated to subclass from the new
    `Aggregation` type, which means that `content` has been replaced with
    `text` and `pattern_id` has been replaced with `type`:

    ```python
    async dev on_match_tag(match: PatternMatch):
       pattern = match.type # instead of match.pattern_id
       text = match.text # instead of match.content
    ```

- `TextFrame` now includes the field `append_to_context` to support setting
  whether or not the encompassing text should be added to the LLM context (by
  the LLM assistant aggregator). It defaults to `True`.

- `TTSService` base class updates:

  - `TTSService`s now accept a new `skip_aggregator_types` to avoid speaking
    certain aggregation types (now determined/returned by the aggregator)

  - Introduced the ability to do a just-in-time transform of text before it gets
    sent to the TTS service via callbacks you can set up via a new init field,
    `text_transforms` or a new method `add_text_transformer()`. This makes it
    possible to do things like introduce TTS-specific tags for spelling or
    emotion or change the pronunciation of something on the
    fly. `remove_text_transformer` has also been added to support removing a
    registered transform callback.

  - TTS services push `AggregatedTextFrame` in addition to `TTSTextFrame`s when
    either an aggregation occurs that should not be spoken or when the TTS
    service supports word-by-word timestamping. In the latter case, the
    `TTSService` preliminarily generates an `AggregatedTextFrame`, aggregated by
    sentence to generate the full sentence content as early as possible.

- Updated `CartesiaTTSService`:

  - Modified use of custom default text_aggregator to avoid deprecation warnings
    and push users towards use of transformers or the `LLMTextProcessor`

  - Added convenience methods for taking advantage of Cartesia's SSML tags:
    spell, emotion, pauses, volume, and speed.

- Updated `RimeTTSService`:

  - Modified use of custom default text_aggregator to avoid deprecation warnings
    and push users towards use of transformers or the `LLMTextProcessor`

  - Added convenience methods for taking advantage of Rime's customization
    options: spell, pauses, pronunciations, and inline speed control.

### Deprecated

- The TTS constructor field, `text_aggregator` is deprecated in favor of the new
  `LLMTextProcessor`. TTSServices still have an internal aggregator for support
  of default behavior, but if you want to override the aggregation behavior, you
  should use the new processor.

- The RTVI `bot-transcription` event is deprecated in favor of the new
  `bot-output` message which is the canonical representation of bot output
  (spoken or not). The code still emits a transcription message for backwards
  compatibility while transition occurs.

- Deprecated `add_pattern_pair` in the `PatternPairAggregator` which takes a
  `pattern_id` and `remove_match` field in favor of the new `add_pattern` method
  which takes a `type` and an `action`

- `english_normalization` input parameter for `MiniMaxHttpTTSService` is
  deprecated, use `test_normalization` instead.

### Fixed

- Fixed an issue in `AWSBedrockLLMService` where the `aws_region` arg was
  always set to `us-east-1` when providing an AWS_REGION env var.

- Fixed an issue with `DeepgramFluxSTTService` where it sometimes failed to reconnect.

- Fixed an issue in `ElevenLabsRealtimeSTTService` where dynamic language
  updates were not working.

- Fixed an issue in `ElevenLabsRealtimeSTTService` where setting the sample
  rate would result in transcripts failing.

- Fixed `InworldTTSService` audio config payload to use camelCase keys expected
  by the Inworld API.

## [0.0.95] - 2025-11-18

### Added

- Added ai-coustics integrated VAD (`AICVADAnalyzer`) with `AICFilter` factory and
  example wiring; leverages the enhancement model for robust detection with no
  ONNX dependency or added processing complexity.

- Added a watchdog to `DeepgramFluxSTTService` to prevent dangling tasks in case the
  user was speaking and we stop receiving audio.

- Introduced a minimum confidence parameter in `DeepgramFluxSTTService` to avoid
  generating transcriptions below a defined threshold.

- Added `ElevenLabsRealtimeSTTService` which implements the Realtime STT
  service from ElevenLabs.

- Added word-level timestamps support to Hume TTS service

### Changed

- ⚠️ Breaking change: `LLMContext.create_image_message()`,
  `LLMContext.create_audio_message()`, `LLMContext.add_image_frame_message()`
  and `LLMContext.add_audio_frames_message()` are now async methods. This fixes
  an issue where the asyncio event loop would be blocked while encoding audio or
  images.

- `ConsumerProcessor` now queues frames from the producer internally instead of
  pushing them directly. This allows us to subclass consumer processors and
  manipulate frames before they are pushed.

- `BaseTextFilter` only require subclasses to implement the `filter()` method.

- Extracted the logic for retrying connections, and create a new `send_with_retry`
  method inside `WebSocketService`.

- Refactored `DeepgramFluxSTTService` to automatically reconnect if sending a
  message fails.

- Updated all STT and TTS services to use consistent error handling pattern with
  `push_error()` method for better pipeline error event integration.

- Added support for `maybe_capture_participant_camera()` and
  `maybe_capture_participant_screen()` for `SmallWebRTCTransport` in the runner
  utils.

- Added Hindi support for Rime TTS services.

- Updated `GeminiTTSService` to use Google Cloud Text-to-Speech streaming API
  instead of the deprecated Gemini API. Now uses `credentials` /
  `credentials_path` for authentication. The `api_key` parameter is deprecated.
  Also, added support for `prompt` parameter for style instructions and
  expressive markup tags. Significantly improved latency with streaming
  synthesis.

- Updated language mappings for the Google and Gemini TTS services to match
  official documentation.

### Deprecated

- The `api_key` parameter in `GeminiTTSService` is deprecated. Use
  `credentials` or `credentials_path` instead for Google Cloud authentication.

### Fixed

- Fixed a `SimliVideoService` connection issue.

- Fixed an issue in the `Runner` where, when using `SmallWebRTCTransport`, the
  `request_data` was not being passed to the `SmallWebRTCRunnerArguments` body.

- Fixed subtle issue of assistant context messages ending up with double spaces
  between words or sentences.

- Fixed an issue where `NeuphonicTTSService` wasn't pushing `TTSTextFrame`s,
  meaning assistant messages weren't being written to context.

- Fixed an issue with OpenTelemetry where tracing wasn't correctly displaying
  LLM completions and tools when using the universal `LLMContext`.

- Fixed issue where `DeepgramFluxSTTService` failed to connect if passing a
  `keyterm` or `tag` containing a space.

- Prevented `HeyGenVideoService` from automatically disconnecting after 5 minutes.

## [0.0.94] - 2025-11-10

### Changed

- Added support for retrying `SpeechmaticsTTSService` when it returns a 503
  error. Default values in `InputParams`.

### Deprecated

- The `KrispFilter` is deprecated and will be removed in a future version. Use
  the `KrispVivaFilter` instead.

### Removed

- `LivekitFrameSerializer` has been removed. Use `LiveKitTransport` instead.

### Fixed

- Fixed a bug related to `LLMAssistantAggregator` where spaces were sometimes
  missing from assistant messages in context.

## [0.0.93] - 2025-11-07

### Added

- Added support for Sarvam Speech-to-Text service (`SarvamSTTService`) with
  streaming WebSocket support for `saarika` (STT) and `saaras` (STT-translate)
  models.

- Added support for passing in a `ToolsSchema` in lieu of a list of provider-
  specific dicts when initializing `OpenAIRealtimeLLMService` or when updating
  it using `LLMUpdateSettingsFrame`.

- Added `TransportParams.audio_out_silence_secs`, which specifies how many
  seconds of silence to output when an `EndFrame` reaches the output
  transport. This can help ensure that all audio data is fully delivered to
  clients.

- Added new `FrameProcessor.broadcast_frame()` method. This will push two
  instances of a given frame class, one upstream and the other downstream.

  ```python
  await self.broadcast_frame(UserSpeakingFrame)
  ```

- Added `MetricsLogObserver` for logging performance metrics from `MetricsFrame`
  instances. Supports filtering via `include_metrics` parameter to control which
  metrics types are logged (TTFB, processing time, LLM token usage, TTS usage,
  smart turn metrics).

- Added `pronunciation_dictionary_locators` to `ElevenLabsTTSService` and
  `ElevenLabsHttpTTSService`.

- Added support for loading external observers. You can now register custom
  pipeline observers by setting the `PIPECAT_OBSERVER_FILES` environment
  variable. This variable should contain a colon-separated list of Python files
  (e.g. `export PIPECAT_OBSERVER_FILES="observer1.py:observer2.py:..."`). Each
  file must define a function with the following signature:

  ```python
  async def create_observers(task: PipelineTask) -> Iterable[BaseObserver]: ...
  ```

- Added support for new sonic-3 languages in `CartesiaTTSService` and
  `CartesiaHttpTTSService`.

- `EndFrame` and `EndTaskFrame` have an optional `reason` field to indicate why
  the pipeline is being ended.

- `CancelFrame` and `CancelTaskFrame` have an optional `reason` field to
  indicate why the pipeline is being canceled. This can be also specified when
  you cancel a task with `PipelineTask.cancel(reason="cancellation reason")`.

- Added `include_prob_metrics` parameter to Whisper STT services to enable access
  to probability metrics from transcription results.

- Added utility functions `extract_whisper_probability()`,
  `extract_openai_gpt4o_probability()`, and `extract_deepgram_probability()` to
  extract probability metrics from `TranscriptionFrame` objects for Whisper-based,
  OpenAI GPT-4o-transcribe, and Deepgram STT services respectively.

- Added `LLMSwitcher.register_direct_function()`. It works much like
  `LLMSwitcher.register_function()` in that it's a shorthand for registering
  functions on all LLMs in the switcher, but for direct functions.

- Added `LLMSwitcher.register_direct_function()`. It works much like
  `LLMSwitcher.register_function()` in that it's a shorthand for registering
  a function on all LLMs in the switcher, except this new method takes a direct
  function (a `FunctionSchema`-less function).

- Added `MCPClient.get_tools_schema()` and `MCPClient.register_tools_schema()`
  as a two-step alternative to `MCPClient.register_tools()`, to allow users to
  pass MCP tools to, say, `GeminiLiveLLMService` (as well as other
  speech-to-speech services) in the constructor.

- Added support for passing in an `LLMSwicher` to `MCPClient.register_tools()`
  (as well as the new `MCPClient.register_tools_schema()`).

- Added `cpu_count` parameter to `LocalSmartTurnAnalyzerV3`. This is set to `1`
  by default for more predictable performance on low-CPU systems.

### Changed

- Updated `simli-ai` to 0.1.25.

- `STTMuteFilter` no longer sends `STTMuteFrame` to the STT service. The filter
  now blocks frames locally without instructing the STT service to stop
  processing audio. This prevents inactivity-related errors (such as 409 errors
  from Google STT) while maintaining the same muting behavior at the application
  level. Important: The STTMuteFilter should be placed _after_ the STT service
  itself.

- Improved `GoogleSTTService` error handling to properly catch gRPC `Aborted`
  exceptions (corresponding to 409 errors) caused by stream inactivity. These
  exceptions are now logged at DEBUG level instead of ERROR level, since they
  indicate expected behavior when no audio is sent for 10+ seconds (e.g., during
  long silences or when audio input is blocked). The service automatically
  reconnects when this occurs.

- Bumped the `fastapi` dependency's upperbound to `<0.122.0`.

- Updated the default model for `GoogleVertexLLMService` to `gemini-2.5-flash`.

- Updated the `GoogleVertexLLMService` to use the `GoogleLLMService` as a base
  class instead of the `OpenAILLMService`.

- Updated STT and TTS services to pass through unverified language codes with a
  warning instead of returning None. This allows developers to use newly
  supported languages before Pipecat's service classes are updated, while still
  providing guidance on verified languages.

### Removed

- Removed `needs_mcp_alternate_schema()` from `LLMService`. The mechanism that
  relied on it went away.

### Fixed

- Restore backwards compatibility for vision/image features (broken in 0.0.92)
  when using non-universal context and assistant aggregators.

- Fixed `DeepgramSTTService._disconnect()` to properly await `is_connected()`
  method call, which is an async coroutine in the Deepgram SDK.

- Fixed an issue where the `SmallWebRTCRequest` dataclass in runner would scrub
  arbitrary request data from client due to camelCase typing. This fixes data
  passthrough for JS clients where `APIRequest` is used.

- Fixed a bug in `GeminiLiveLLMService` where in some circumstances it wouldn't
  respond after a tool call.

- Fixed `GeminiLiveLLMService` session resumption after a connection timeout.

- `GeminiLiveLLMService` now properly supports context-provided system
  instruction and tools.

- Fixed `GoogleLLMService` token counting to avoid double-counting tokens when
  Gemini sends usage metadata across multiple streaming chunks.

## [0.0.92] - 2025-10-31 🎃 "The Haunted Edition" 👻

### Added

- Added a new `DeepgramHttpTTSService`, which delivers a meaningful reduction
  in latency when compared to the `DeepgramTTSService`.

- Add support for `speaking_rate` input parameter in `GoogleHttpTTSService`.

- Added `enable_speaker_diarization` and `enable_language_identification` to
  `SonioxSTTService`.

- Added `SpeechmaticsTTSService`, which uses Speechmatic's TTS API. Updated
  examples 07a\* to use the new TTS service.

- Added support for including images or audio to LLM context messages using
  `LLMContext.create_image_message()` or `LLMContext.create_image_url_message()`
  (not all LLMs support URLs) and `LLMContext.create_audio_message()`. For
  example, when creating `LLMMessagesAppendFrame`:

  ```python
  message = LLMContext.create_image_message(image=..., size=...)
  await self.push_frame(LLMMessagesAppendFrame(messages=[message], run_llm=True))
  ```

- New event handlers for the `DeepgramFluxSTTService`: `on_start_of_turn`,
  `on_turn_resumed`, `on_end_of_turn`, `on_eager_end_of_turn`, `on_update`.

- Added `generation_config` parameter support to `CartesiaTTSService` and
  `CartesiaHttpTTSService` for Cartesia Sonic-3 models. Includes a new
  `GenerationConfig` class with `volume` (0.5-2.0), `speed` (0.6-1.5),
  and `emotion` (60+ options) parameters for fine-grained speech generation
  control.

- Expanded support for univeral `LLMContext` to `OpenAIRealtimeLLMService`.
  As a reminder, the context-setup pattern when using `LLMContext` is:

  ```python
  context = LLMContext(messages, tools)
  context_aggregator = LLMContextAggregatorPair(context)
  ```

  (Note that even though `OpenAIRealtimeLLMService` now supports the universal
  `LLMContext`, it is not meant to be swapped out for another LLM service at
  runtime with `LLMSwitcher`.)

  Note: `TranscriptionFrame`s and `InterimTranscriptionFrame`s now go upstream
  from `OpenAIRealtimeLLMService`, so if you're using `TranscriptProcessor`,
  say, you'll want to adjust accordingly:

  ```python
  pipeline = Pipeline(
      [
          transport.input(),
          context_aggregator.user(),
          # BEFORE
          llm,
          transcript.user(),
          # AFTER
          transcript.user(),
          llm,
          transport.output(),
          transcript.assistant(),
          context_aggregator.assistant(),
      ]
  )
  ```

  Also worth noting: whether or not you use the new context-setup pattern with
  `OpenAIRealtimeLLMService`, some types have changed under the hood:

  ```python
  ## BEFORE:

  # Context aggregator type
  context_aggregator: OpenAIContextAggregatorPair

  # Context frame type
  frame: OpenAILLMContextFrame

  # Context type
  context: OpenAIRealtimeLLMContext
  # or
  context: OpenAILLMContext

  ## AFTER:

  # Context aggregator type
  context_aggregator: LLMContextAggregatorPair

  # Context frame type
  frame: LLMContextFrame

  # Context type
  context: LLMContext
  ```

  Also note that `RealtimeMessagesUpdateFrame` and
  `RealtimeFunctionCallResultFrame` have been deprecated, since they're no
  longer used by `OpenAIRealtimeLLMService`. OpenAI Realtime now works more
  like other LLM services in Pipecat, relying on updates to its context, pushed
  by context aggregators, to update its internal state. Listen for
  `LLMContextFrame`s for context updates.

  Finally, `LLMTextFrame`s are no longer pushed from `OpenAIRealtimeLLMService`
  when it's configured with `output_modalities=['audio']`. If you need
  to process its output, listen for `TTSTextFrame`s instead.

- Expanded support for universal `LLMContext` to `GeminiLiveLLMService`.
  As a reminder, the context-setup pattern when using `LLMContext` is:

  ```python
  context = LLMContext(messages, tools)
  context_aggregator = LLMContextAggregatorPair(context)
  ```

  (Note that even though `GeminiLiveLLMService` now supports the universal
  `LLMContext`, it is not meant to be swapped out for another LLM service at
  runtime with `LLMSwitcher`.)

  Worth noting: whether or not you use the new context-setup pattern with
  `GeminiLiveLLMService`, some types have changed under the hood:

  ```python
  ## BEFORE:

  # Context aggregator type
  context_aggregator: GeminiLiveContextAggregatorPair

  # Context frame type
  frame: OpenAILLMContextFrame

  # Context type
  context: GeminiLiveLLMContext
  # or
  context: OpenAILLMContext

  ## AFTER:

  # Context aggregator type
  context_aggregator: LLMContextAggregatorPair

  # Context frame type
  frame: LLMContextFrame

  # Context type
  context: LLMContext
  ```

  Also note that `LLMTextFrame`s are no longer pushed from `GeminiLiveLLMService`
  when it's configured with `modalities=GeminiModalities.AUDIO`. If you need
  to process its output, listen for `TTSTextFrame`s instead.

### Changed

- The development runner's `/start` endpoint now supports passing
  `dailyRoomProperties` and `dailyMeetingTokenProperties` in the request body
  when `createDailyRoom` is true. Properties are validated against the
  `DailyRoomProperties` and `DailyMeetingTokenProperties` types respectively
  and passed to Daily's room and token creation APIs.

- `UserImageRawFrame` new fields `append_to_context` and `text`. The
  `append_to_context` field indicates if this image and text should be added to
  the LLM context (by the LLM assistant aggregator). The `text` field, if set,
  might also guide the LLM or the vision service on how to analyze the image.

- `UserImageRequestFrame` new fiels `append_to_context` and `text`. Both fields
  will be used to set the same fields on the captured `UserImageRawFrame`.

- `UserImageRequestFrame` don't require function call name and ID anymore.

- Updated `MoondreamService` to process `UserImageRawFrame`.

- `VisionService` expects `UserImageRawFrame` in order to analyze images.

- `DailyTransport` triggers `on_error` event if transcription can't be started
  or stopped.

- `DailyTransport` updates: `start_dialout()` now returns two values:
  `session_id` and `error`. `start_recording()` now returns two values:
  `stream_id` and `error`.

- Updated `daily-python` to 0.21.0.

- `SimliVideoService` now accepts `api_key` and `face_id` parameters directly,
  with optional `params` for `max_session_length` and `max_idle_time`
  configuration, aligning with other Pipecat service patterns.

- Updated the default model to `sonic-3` for `CartesiaTTSService` and
  `CartesiaHttpTTSService`.

- `FunctionFilter` now has a `filter_system_frames` arg, which controls whether
  or not SystemFrames are filtered.

- Upgraded `aws_sdk_bedrock_runtime` to v0.1.1 to resolve potential CPU issues
  when running `AWSNovaSonicLLMService`.

### Deprecated

- The `expect_stripped_words` parameter of `LLMAssistantAggregatorParams` is
  ignored when used with the newer `LLMAssistantAggregator`, which now handles
  word spacing automatically.

- `LLMService.request_image_frame()` is deprecated, push a
  `UserImageRequestFrame` instead.

- `UserResponseAggregator` is deprecated and will be removed in a future version.

- The `send_transcription_frames` argument to `OpenAIRealtimeLLMService` is
  deprecated. Transcription frames are now always sent. They go upstream, to be
  handled by the user context aggregator. See "Added" section for details.

- Types in `pipecat.services.openai.realtime.context` and
  `pipecat.services.openai.realtime.frames` are deprecated, as they're no
  longer used by `OpenAIRealtimeLLMService`. See "Added" section for details.

- `SimliVideoService` `simli_config` parameter is deprecated. Use `api_key` and
  `face_id` parameters instead.

### Removed

- Removed `enable_non_final_tokens` and `max_non_final_tokens_duration_ms` from
  `SonioxSTTService`.

- Removed the `aiohttp_session` arg from `SarvamTTSService` as it's no longer
  used.

### Fixed

- Fixed a `PipelineTask` issue that was causing an idle timeout for frames that
  were being generated but not reaching the end of the pipeline. Since the exact
  point when frames are discarded is unknown, we now monitor pipeline frames
  using an observer. If the observer detects frames are being generated, it will
  prevent the pipeline from being considered idle.

- Fixed an issue in `HumeTTSService` that was only using Octave 2, which does
  not support the `description` field. Now, if a description is provided, it
  switches to Octave 1.

- Fixed an issue where `DailyTransport` would timeout prematurely on join and on
  leave.

- Fixed an issue in the runner where starting a DailyTransport room via
  `/start` didn't support using the `DAILY_SAMPLE_ROOM_URL` env var.

- Fixed an issue in `ServiceSwitcher` where the `STTService`s would result in
  all STT services producing `TranscriptionFrame`s.

### Other

- Updated all vision 12-series foundational examples to load images from a file.

- Added 14-series video examples for different services. These new examples
  request an image from the user camera through a function call.

## [0.0.91] - 2025-10-21

### Added

- It is now possible to start a bot from the `/start` endpoint when using the
  runner Daily's transport. This follows the Pipecat Cloud format with
  `createDailyRoom` and `body` fields in the POST request body.

- Added an ellipsis character (`…`) to the end of sentence detection in the
  string utils.

- Expanded support for universal `LLMContext` to `AWSNovaSonicLLMService`.
  As a reminder, the context-setup pattern when using `LLMContext` is:

  ```python
  context = LLMContext(messages, tools)
  context_aggregator = LLMContextAggregatorPair(context)
  ```

  (Note that even though `AWSNovaSonicLLMService` now supports the universal
  `LLMContext`, it is not meant to be swapped out for another LLM service at
  runtime with `LLMSwitcher`.)

  Worth noting: whether or not you use the new context-setup pattern with
  `AWSNovaSonicLLMService`, some types have changed under the hood:

  ```python
  ## BEFORE:

  # Context aggregator type
  context_aggregator: AWSNovaSonicContextAggregatorPair

  # Context frame type
  frame: OpenAILLMContextFrame

  # Context type
  context: AWSNovaSonicLLMContext
  # or
  context: OpenAILLMContext

  ## AFTER:

  # Context aggregator type
  context_aggregator: LLMContextAggregatorPair

  # Context frame type
  frame: LLMContextFrame

  # Context type
  context: LLMContext
  ```

- Added support for `bulbul:v3` model in `SarvamTTSService` and
  `SarvamHttpTTSService`.

- Added `keyterms_prompt` parameter to `AssemblyAIConnectionParams`.

- Added `speech_model` parameter to `AssemblyAIConnectionParams` to access the
  multilingual model.

- Added support for trickle ICE to the `SmallWebRTCTransport`.

- Added support for updating `OpenAITTSService` settings (`instructions` and
  `speed`) at runtime via `TTSUpdateSettingsFrame`.

- Added `--whatsapp` flag to runner to better surface WhatsApp transport logs.

- Added `on_connected` and `on_disconnected` events to TTS and STT
  websocket-based services.

- Added an `aggregate_sentences` arg in `ElevenLabsHttpTTSService`, where the
  default value is True.

- Added a `room_properties` arg to the Daily runner's `configure()` method,
  allowing `DailyRoomProperties` to be provided.

- The runner `--folder` argument now supports downloading files from
  subdirectories.

### Changed

- `RunnerArguments` now include the `body` field, so there's no need to add it
  to subclasses. Also, all `RunnerArguments` fields are now keyword-only.

- `CartesiaSTTService` now inherits from `WebsocketSTTService`.

- Package upgrades:

  - `daily-python` upgraded to 0.20.0.
  - `openai` upgraded to support up to 2.x.x.
  - `openpipe` upgraded to support up to 5.x.x.

- `SpeechmaticsSTTService` updated dependencies for `speechmatics-rt>=0.5.0`.

### Deprecated

- The `send_transcription_frames` argument to `AWSNovaSonicLLMService` is
  deprecated. Transcription frames are now always sent. They go upstream, to be
  handled by the user context aggregator. See "Added" section for details.

- Types in `pipecat.services.aws.nova_sonic.context` are deprecated, as they're
  no longer used by `AWSNovaSonicLLMService`. See "Added" section for
  details.

### Fixed

- Fixed an issue where the `RTVIProcessor` was sending duplicate
  `UserStartedSpeakingFrame` and `UserStoppedSpeakingFrame` messages.

- Fixed an issue in `AWSBedrockLLMService` where both `temperature` and `top_p`
  were always sent together, causing conflicts with models like Claude Sonnet 4.5
  that don't allow both parameters simultaneously. The service now only includes
  inference parameters that are explicitly set, and `InputParams` defaults have
  been changed to `None` to rely on AWS Bedrock's built-in model defaults.

- Fixed an issue in `RivaSegmentedSTTService` where a runtime error occurred due
  to a mismatch in the `_handle_transcription` method's signature.

- Fixed multiple pipeline task cancellation issues. `asyncio.CancelledError` is
  now handled properly in `PipelineTask` making it possible to cancel an asyncio
  task that it's executing a `PipelineRunner` cleanly. Also,
  `PipelineTask.cancel()` does not block anymore waiting for the `CancelFrame`
  to reach the end of the pipeline (going back to the behavior in < 0.0.83).

- Fixed an issue in `ElevenLabsTTSService` and `ElevenLabsHttpTTSService` where
  the Flash models would split words, resulting in a space being inserted
  between words.

- Fixed an issue where audio filters' `stop()` would not be called when using
  `CancelFrame`.

- Fixed an issue in `ElevenLabsHttpTTSService`, where
  `apply_text_normalization` was incorrectly set as a query parameter. It's now
  being added as a request parameter.

- Fixed an issue where `RimeHttpTTSService` and `PiperTTSService` could generate
  incorrectly 16-bit aligned audio frames, potentially leading to internal
  errors or static audio.

- Fixed an issue in `SpeechmaticsSTTService` where `AdditionalVocabEntry` items
  needed to have `sounds_like` for the session to start.

### Other

- Added foundational example `47-sentry-metrics.py`, demonstrating how to use the
  `SentryMetrics` processor.

- Added foundational example `14x-function-calling-openpipe.py`.

## [0.0.90] - 2025-10-10

### Added

- Added audio filter `KrispVivaFilter` using the Krisp VIVA SDK.

- Added `--folder` argument to the runner, allowing files saved in that folder
  to be downloaded from `http://HOST:PORT/file/FILE`.

- Added `GeminiLiveVertexLLMService`, for accessing Gemini Live via Google
  Vertex AI.

- Added some new configuration options to `GeminiLiveLLMService`:

  - `thinking`
  - `enable_affective_dialog`
  - `proactivity`

  Note that these new configuration options require using a newer model than
  the default, like "gemini-2.5-flash-native-audio-preview-09-2025". The last
  two require specifying `http_options=HttpOptions(api_version="v1alpha")`.

- Added `on_pipeline_error` event to `PipelineTask`. This event will get fired
  when an `ErrorFrame` is pushed (use `FrameProcessor.push_error()`).

  ```python
  @task.event_handler("on_pipeline_error")
  async def on_pipeline_error(task: PipelineTask, frame: ErrorFrame): ...
  ```

- Added a `service_tier` `InputParam` to the `BaseOpenAILLMService`. This
  parameter can influence the latency of the response. For example `"priority"`
  will result in faster completions, but in exchange for a higher price.

### Changed

- Updated `GeminiLiveLLMService` to use the `google-genai` library rather than
  use WebSockets directly.

### Deprecated

- `LivekitFrameSerializer` is now deprecated. Use `LiveKitTransport` instead.

- `pipecat.service.openai_realtime` is now deprecated, use
  `pipecat.services.openai.realtime` instead or
  `pipecat.services.azure.realtime` for Azure Realtime.

- `pipecat.service.aws_nova_sonic` is now deprecated, use
  `pipecat.services.aws.nova_sonic` instead.

- `GeminiMultimodalLiveLLMService` is now deprecated, use
  `GeminiLiveLLMService`.

### Fixed

- Fixed a `GoogleVertexLLMService` issue that would generate an error if no
  token information was returned.

- `GeminiLiveLLMService` will now end gracefully (i.e. after the bot has
  finished) upon receiving an `EndFrame`.

- `GeminiLiveLLMService` will try to seamlessly reconnect when it loses its
  connection.

## [0.0.89] - 2025-10-07

### Fixed

- Reverted a change introduced in 0.0.88 that was causing pipelines to be frozen
  when using interruption strategies and processors that block interruption
  frames (e.g. `STTMuteFilter`).

## [0.0.88] - 2025-10-07

### Added

- Added support for Nano Banana models to `GoogleLLMService`. For example, you
  can now use the `gemini-2.5-flash-image` model to generate images.

- Added `HumeTTSService` for text-to-speech synthesis using Hume AI's expressive
  voice models. Provides high-quality, emotionally expressive speech synthesis
  with support for various voice models. Includes example in
  `examples/foundational/07ad-interruptible-hume.py`. Use with:
  `uv pip install pipecat-ai[hume]`.

### Changed

- Updated default `GoogleLLMService` model to `gemini-2.5-flash`.

### Deprecated

- PlayHT is shutting down their API on December 31st, 2025. As a result,
  `PlayHTTTSService` and `PlayHTHttpTTSService` are deprecated and will be
  removed in a future version.

### Fixed

- Fixed an issue with `AWSNovaSonicLLMService` where the client wouldn't
  connect due to a breaking change in the AWS dependency chain.

- `PermissionError` is now caught if NLTK's `punkt_tab` can't be downloaded.

- Fixed an issue that would cause wrong user/assistant context ordering when
  using interruption strategies.

- Fixed RTVI incoming message handling, broken in 0.0.87.

## [0.0.87] - 2025-10-02

### Added

- Added `WebsocketSTTService` base class for websocket-based STT services.
  Combines STT functionality with websocket connectivity, providing automatic
  error handling and reconnection capabilities with exponential backoff.

- Added `DeepgramFluxSTTService` for real-time speech recognition using
  Deepgram's Flux WebSocket API. Flux understands conversational flow and
  automatically handles turn-taking.

- Added RTVI messages for user/bot audio levels and system logs.

- Include OpenAI-based LLM services cached tokens to `MetricsFrame`.

### Changed

- Updated the default model for `AnthropicLLMService` to
  `claude-sonnet-4-5-20250929`.

### Deprecated

- `DailyTransportMessageFrame` and `DailyTransportMessageUrgentFrame` are
  deprecated, use `DailyOutputTransportMessageFrame` and
  `DailyOutputTransportMessageUrgentFrame` respectively instead.

- `LiveKitTransportMessageFrame` and `LiveKitTransportMessageUrgentFrame` are
  deprecated, use `LiveKitOutputTransportMessageFrame` and
  `LiveKitOutputTransportMessageUrgentFrame` respectively instead.

- `TransportMessageFrame` and `TransportMessageUrgentFrame` are deprecated, use
  `OutputTransportMessageFrame` and `OutputTransportMessageUrgentFrame`
  respectively instead.

- `InputTransportMessageUrgentFrame` is deprecated, use
  `InputTransportMessageFrame` instead.

- `DailyUpdateRemoteParticipantsFrame` is deprecated and will be removed in a
  future version. Instead, create your own custom frame and handle it in the
  `@transport.output().event_handler("on_after_push_frame")` event handler or a
  custom processor.

### Fixed

- Fixed an issue in `AWSBedrockLLMService` where timeout exceptions weren't
  being detected.

- Fixed a `PipelineTask` issue that could prevent the application to exit if
  `task.cancel()` was called when the task was already finished.

- Fixed an issue where local SmartTurn was not being ran in a separate thread.

## [0.0.86] - 2025-09-24

### Added

- Added `HeyGenTransport`. This is an integration for HeyGen Interactive
  Avatar. A video service that handles audio streaming and requests HeyGen to
  generate avatar video responses. (see https://www.heygen.com/). When used, the
  Pipecat bot joins the same virtual room as the HeyGen Avatar and the user.

- Added support to `TwilioFrameSerializer` for `region` and `edge` settings.

- Added support for using universal `LLMContext` with:

  - `LLMLogObserver`
  - `GatedLLMContextAggregator` (formerly `GatedOpenAILLMContextAggregator`)
  - `LangchainProcessor`
  - `Mem0MemoryService`

- Added `StrandsAgentProcessor` which allows you to use the Strands Agents
  framework to build your voice agents.
  See https://strandsagents.com

- Added `ElevenLabsSTTService` for speech-to-text transcription.

- Added a peer connection monitor to the `SmallWebRTCConnection` that
  automatically disconnects if the connection fails to establish within
  the timeout (1 minute by default).

- Added memory cleanup improvements to reduce memory peaks.

- Added `on_before_process_frame`, `on_after_process_frame`,
  `on_before_push_frame` and `on_after_push_frame`. These are synchronous events
  that get called before and after a frame is processed or pushed. Note that
  these events are synchrnous so they should ideally perform lightweight tasks
  in order to not block the pipeline. See
  `examples/foundational/45-before-and-after-events.py`.

- Added `on_before_leave` synchronous event to `DailyTransport`.

- Added `on_before_disconnect` synchronous event to `LiveKitTransport`.

- It is now possible to register synchronous event handlers. By default, all
  event handlers are executed in a separate task. However, in some cases we want
  to guarantee order of execution, for example, executing something before
  disconnecting a transport.

  ```python
  self._register_event_handler("on_event_name", sync=True)
  ```

- Added support for global location in `GoogleVertexLLMService`. The service now
  supports both regional locations (e.g., "us-east4") and the "global" location
  for Vertex AI endpoints. When using "global" location, the service will use
  `aiplatform.googleapis.com` as the API host instead of the regional format.

- Added `on_pipeline_finished` event to `PipelineTask`. This event will get
  fired when the pipeline is done running. This can be the result of a
  `StopFrame`, `CancelFrame` or `EndFrame`.

  ```python
  @task.event_handler("on_pipeline_finished")
  async def on_pipeline_finished(task: PipelineTask, frame: Frame): ...
  ```

- Added support for new RTVI `send-text` event, along with the ability to toggle
  the audio response off (skip tts) while handling the new context.

### Changed

- Updated `aiortc` to 1.13.0.

- Updated `sentry` to 2.38.0.

- `BaseOutputTransport` methods `write_audio_frame` and `write_video_frame` now
  return a boolean to indicate if the transport implementation was able to write
  the given frame or not.

- Updated Silero VAD model to v6.

- Updated `livekit` to 1.0.13.

- `torch` and `torchaudio` are no longer required for running Smart Turn
  locally. This avoids gigabytes of dependencies being installed.

- Updated `websockets` dependency to support version 15.0. Removed deprecated
  usage of `ConnectionClosed.code` and `ConnectionClosed.reason` attributes in
  `AWSTranscribeSTTService` for compatibility.

- Refactored `pyproject.toml` to reduce websockets dependency repetition using
  self-referencing extras. All websockets-dependent services now reference a
  shared `websockets-base` extra.

### Deprecated

- `GladiaSTTService`'s `confidence` arg is deprecated. `confidence` is no
  longer needed to determine which transcription or translation frames to
  emit.

- `PipelineTask` events `on_pipeline_stopped`, `on_pipeline_ended` and
  `on_pipeline_cancelled` are now deprecated. Use `on_pipeline_finished`
  instead.

- Support for the RTVI `append-to-context` event, in lieu of the new `send-text`
  event and making way for future events like `send-image`.

### Fixed

- Fixed an issue where the pipeline could freeze if a task cancellation never
  completed because a third-party library swallowed asyncio.CancelledError. We
  now apply a timeout to task cancellations to prevent these freezes. If the
  timeout is reached, the system logs warnings and leaves dangling tasks behind,
  which can help diagnose where cancellation is being blocked.

- Fixed an `AudioBufferProcessor` issues that was causing user audio to be
  missing in stereo recordings causing bot and user overlaps.

- Fixed a `BaseOutputTransport` issue that could produce large saved
  `AudioBufferProcessor` files when using an audio mixer.

- Fixed a `PipelineRunner` issue on Windows where setting up SIGINT and SIGTERM
  was raising an exception.

- Fixed an issue where multiple handlers for an event would not run in parallel.

- Fixed `DailyTransport.sip_call_transfer()` to automatically use the session
  ID from the `on_dialin_connected` event, when not explicitly provided. Now
  supports cold transfers (from incoming dial-in calls) by automatically
  tracking session IDs from connection events.

- Fixed a memory leak in `SmallWebRTCTransport`. In `aiortc`, when you receive
  a `MediaStreamTrack` (audio or video), frames are produced asynchronously. If
  the code never consumes these frames, they are queued in memory, causing a
  memory leak.

- Fixed an issue in `AsyncAITTSService`, where `TTSTextFrames` were not being
  pushed.

- Fixed an issue that would cause `push_interruption_task_frame_and_wait()` to
  not wait if a previous interruption had already happened.

- Fixed a couple of bugs in `ServiceSwitcher`:

  - Using multiple `ServiceSwitcher`s in a pipeline would result in an error.
  - `ServiceSwitcherFrame`s (such as `ManuallySwitchServiceFrame`s) were having
    an effect too early, essentially "jumping the queue" in terms of pipeline
    frame ordering.

- Fixed a self-cancellation deadlock in `UserIdleProcessor` when returning
  `False` from an idle callback. The task now terminates naturally instead of
  attempting to cancel itself.

- Fixed an issue in `AudioBufferProcessor` where a recording is not created
  when a bot speaks and user input is blocked.

- Fixed a `FastAPIWebsocketTransport` and `SmallWebRTCTransport` issue where
  `on_client_disconnected` would be triggered when the bot ends the
  conversation. That is, `on_client_disconnected` should only be triggered when
  the remote client actually disconnects.

- Fixed an issue in `HeyGenVideoService` where the `BotStartedSpeakingFrame`
  was blocked from moving through the Pipeline.

## [0.0.85] - 2025-09-12

### Added

- `AzureSTTService` now pushes interim transcriptions.

- Added `voice_cloning_key` to `GoogleTTSService` to support custom cloned
  voices.

- Added `speaking_rate` to `GoogleTTSService.InputParams` to control the
  speaking rate.

- Added a `speed` arg to `OpenAITTSService` to control the speed of the voice
  response.

- Added `FrameProcessor.push_interruption_task_frame_and_wait()`. Use this
  method to programatically interrupt the bot from any part of the
  pipeline. This guarantees that all the processors in the pipeline are
  interrupted in order (from upstream to downstream). Internally, this works by
  first pushing an `InterruptionTaskFrame` upstream until it reaches the
  pipeline task. The pipeline task then generates an `InterruptionFrame`, which
  flows downstream through all processors. Once the `InterruptionFrame` has
  reaches the processor waiting for the interruption, the function returns and
  execution continues after the call. Think of it as sending an upstream request
  for interruption and waiting until the acknowledgment flows back downstream.

- Added new base `TaskFrame` (which is a system frame). This is the base class
  for all task frames (`EndTaskFrame`, `CancelTaskFrame`, etc.) that are meant
  to be pushed upstream to reach the pipeline task.

- Expanded support for universal `LLMContext` to the AWS Bedrock LLM service.
  Using the universal `LLMContext` and associated `LLMContextAggregatorPair` is
  a pre-requisite for using `LLMSwitcher` to switch between LLMs at runtime.

- Added new fields to the development runner's `parse_telephony_websocket`
  method in support of providing dynamic data to a bot.

  - Twilio: Added a new `body` parameter, which parses the websocket message
    for `customParameters`. Provide data via the `Parameter` nouns in your
    TwiML to use this feature.
  - Telnyx & Exotel: Both providers make the `to` and `from` phone numbers
    available in the websocket messages. You can now access these numbers as
    `call_data["to"]` and `call_data["from"]`.

  Note: Each telephony provider offers different features. Refer to the
  corresponding example in `pipecat-examples` to see how to pass custom data
  to your bot.

- Added `body` to the `WebsocketRunnerArguments` as an optional parameter.
  Custom `body` information can be passed from the server into the bot file via
  the `bot()` method using this new parameter.

- Added video streaming support to `LiveKitTransport`.

- Added `OpenAIRealtimeLLMService` and `AzureRealtimeLLMService` which provide
  access to OpenAI Realtime.

### Changed

- `pipeline.tests.utils.run_test()` now allows passing `PipelineParams` instead
  of individual parameters.

### Removed

- Remove `VisionImageRawFrame` in favor of context frames (`LLMContextFrame` or
  `OpenAILLMContextFrame`).

### Deprecated

- `BotInterruptionFrame` is now deprecated, use `InterruptionTaskFrame` instead.

- `StartInterruptionFrame` is now deprected, use `InterruptionFrame` instead.

- Deprecate `VisionImageFrameAggregator` because `VisionImageRawFrame` has been
  removed. See the `12*` examples for the new recommended replacement pattern.

- `NoisereduceFilter` is now deprecated and will be removed in a future
  version. Use other audio filters like `KrispFilter` or `AICFilter`.

- Deprecated `OpenAIRealtimeBetaLLMService` and `AzureRealtimeBetaLLMService`.
  Use `OpenAIRealtimeLLMService` and `AzureRealtimeLLMService`, respectively.
  Each service will be removed in an upcoming version, 1.0.0.

### Fixed

- Fixed a `BaseOutputTransport` issue that caused incorrect detection of when
  the bot stopped talking while using an audio mixer.

- Fixed a `LiveKitTransport` issue where RTVI messages were not properly
  encoded.

- Add additional fixups to Mistral context messages to ensure they meet
  Mistral-specific requirements, avoiding Mistral "invalid request" errors.

- Fixed `DailyTransport` transcription handling to gracefully handle missing
  `rawResponse` field in transcription messages, preventing KeyError crashes.

## [0.0.84] - 2025-09-05

### Added

- Add the ability to send DTMF to `LiveKitTransport`.

- Expanded support for universal `LLMContext` to the Anthropic LLM service.
  Using the universal `LLMContext` and associated `LLMContextAggregatorPair` is
  a pre-requisite for using `LLMSwitcher` to switch between LLMs at runtime.

### Changed

- Updated `daily-python` to 0.19.9.

- Restored `DailyTransport`'s native DTMF support using Daily's `send_dtmf()`
  method instead of generated audio tones.

### Fixed

- Fixed a `AWSBedrockLLMService` crash caused by an extra `await`.

- Fixed a `OpenAIImageGenService` issue where it was not creating
  `URLImageRawFrame` correctly.

## [0.0.83] - 2025-09-03

### Added

- Added multilingual support for AsyncAI in `AsyncAITTSService` and `AsyncAIHttpTTSService`.

  - New `languages`: `es`, `fr`, `de`, `it`.

- Added new frames `InputTransportMessageUrgentFrame` and
  `DailyInputTransportMessageUrgentFrame` for transport messages received from
  external sources.

- Added `UserSpeakingFrame`. This will be sent upstream and downstream while VAD
  detects the user is speaking.

- Expanded support for universal `LLMContext` to more LLM services. Using the
  universal `LLMContext` and associated `LLMContextAggregatorPair` is a
  pre-requisite for using `LLMSwitcher` to switch between LLMs at runtime.
  Here are the newly-supported services:

  - Azure
  - Cerebras
  - Deepseek
  - Fireworks AI
  - Google Vertex AI
  - Grok
  - Groq
  - Mistral
  - NVIDIA NIM
  - Ollama
  - OpenPipe
  - OpenRouter
  - Perplexity
  - Qwen
  - SambaNova
  - Together.ai

- Added support for WhatsApp User-initiated Calls.

- Added new audio filter `AICFilter`, speech enhancement for improving VAD/STT
  performance, no ONNX dependency.
  See https://ai-coustics.com/sdk/

- Added a timeout around cancel input tasks to prevent indefinite hangs when
  cancellation is swallowed by third-party code.

- Added `pipecat.extensions.ivr` for automated IVR system navigation with
  configurable goals and conversation handling. Supports DTMF input, verbal
  responses, and intelligent menu traversal.

  Basic usage:

  ```python
  from pipecat.extensions.ivr.ivr_navigator import IVRNavigator

  # Create IVR navigator with your goal
  ivr_navigator = IVRNavigator(
      llm=llm_service,
      ivr_prompt="Navigate to billing department to dispute a charge"
  )

  # Handle different outcomes
  @ivr_navigator.event_handler("on_conversation_detected")
  async def on_conversation(processor, conversation_history):
      # Switch to normal conversation mode
      pass

  @ivr_navigator.event_handler("on_ivr_status_changed")
  async def on_ivr_status(processor, status):
      if status == IVRStatus.COMPLETED:
          # End pipeline, transfer call, or start bot conversation
      elif status == IVRStatus.STUCK:
          # Handle navigation failure
  ```

- `BaseOutputTransport` now implements `write_dtmf()` by loading DTMF audio and
  sending it through the transport. This makes sending DTMF generic across all
  output transports.

- Added new config parameters to `GladiaSTTService`.
  - PreProcessingConfig > `audio_enhancer` to enhance audio quality.
  - CustomVocabularyItem > `pronunciations` and `language` to specify special
    pronunciations and in which language it will be pronounced.

### Changed

- `UserStartedSpeakingFrame` and `UserStoppedSpeakingFrame` are also pushed
  upstream.

- `ParallelPipeline` now waits for `CancelFrame` to finish in all branches
  before pushing it downstream.

- Added `sip_codecs` to the `DailyRoomSipParams`.

- Updated the `configure()` function in `pipecat.runner.daily` to include new
  args to create SIP-enabled rooms. Additionally, added new args to control the
  room and token expiration durations.

- `pipecat.frames.frames.KeypadEntry` is deprecated and has been moved to
  `pipecat.audio.dtmf.types.KeypadEntry`.

- Updated `RimeTTSService`'s flush_audio message to conform with Rime's official
  API.

- Updated the default model for `CerebrasLLMService` to GPT-OSS-120B.

### Removed

- Remove `StopInterruptionFrame`. This was a legacy frame that was not being
  used really anywhere and it didn't provide any useful meaning. It was only
  pushed after `UserStoppedSpeakingFrame`, so developers can just use
  `UserStoppedSpeakingFrame`.

- `DailyTransport.write_dtmf()` has been removed in favor of the generic
  `BaseOutputTransport.write_dtmf()`.

- Remove deprecated `DailyTransport.send_dtmf()`.

### Deprecated

- Transports have been re-organized.

  ```
  pipecat.transports.network.small_webrtc        -> pipecat.transports.smallwebrtc.transport
  pipecat.transports.network.webrtc_connection   -> pipecat.transports.smallwebrtc.connection
  pipecat.transports.network.websocket_client    -> pipecat.transports.websocket.client
  pipecat.transports.network.websocket_server    -> pipecat.transports.websocket.server
  pipecat.transports.network.fastapi_websocket   -> pipecat.transports.websocket.fastapi
  pipecat.transports.services.daily              -> pipecat.transports.daily.transport
  pipecat.transports.services.helpers.daily_rest -> pipecat.transports.daily.utils
  pipecat.transports.services.livekit            -> pipecat.transports.livekit.transport
  pipecat.transports.services.tavus              -> pipecat.transports.tavus.transport
  ```

- `pipecat.frames.frames.KeypadEntry` is deprecated use
  `pipecat.audio.dtmf.types.KeypadEntry` instead.

### Fixed

- Fixed an issue where messages received from the transport were always being resent.

- Fixed `SmallWebRTCTransport` to not use `mid` to decide if the transceiver should
  be `sendrecv` or not.

- Fixed an issue where Deepgram swallowed `asyncio.CancelledError` during
  disconnect, preventing tasks from being cancelled.

- Fixed an issue where `PipelineTask` was not cleaning up the observers.

### Performance

- Reduced latency and improved memory performance in `Mem0MemoryService`.

## [0.0.82] - 2025-08-28

### Added

- Added a new `LLMRunFrame` to trigger an LLM response:

  ```python
  await task.queue_frames([LLMRunFrame()])
  ```

  This replaces `OpenAILLMContextFrame`, which you’d previously typically use
  like this:

  ```python
  await task.queue_frames([context_aggregator.user().get_context_frame()])
  ```

  Use this way of kicking off your conversation when you’ve already initialized
  your context and are simply instructing the bot when to go:

  ```python
  context = OpenAILLMContext(messages, tools)
  context_aggregator = llm.create_context_aggregator(context)

  # ...


  @transport.event_handler("on_client_connected")
  async def on_client_connected(transport, client):
      # Kick off the conversation.
      await task.queue_frames([LLMRunFrame()])
  ```

  Note that if you want to add new messages when kicking off the conversation,
  you could use `LLMMessagesAppendFrame` with `run_llm=True` instead:

  ```python
  @transport.event_handler("on_client_connected")
  async def on_client_connected(transport, client):
      # Kick off the conversation.
      await task.queue_frames([LLMMessagesAppendFrame(new_messages, run_llm=True)])
  ```

  In the rare case you don’t have a context aggregator in your pipeline, then
  you may continue using a context frame.

- Added support for switching between audio+text to text-only modes within the
  same pipeline. This is done by pushing
  `LLMConfigureOutputFrame(skip_tts=True)` to enter text-only mode, and
  disabling it to return to audio+text. The LLM will still generate tokens and
  add them to the context, but they will not be sent to TTS.

- Added `skip_tts` field to `TextFrame`. This lets a text frame bypass TTS while
  still being included in the LLM context. Useful for cases like structured text
  that isn’t meant to be spoken but should still contribute to context.

- Added a `cancel_timeout_secs` argument to `PipelineTask` which defines how
  long the pipeline has to complete cancellation. When `PipelineTask.cancel()`
  is called, a `CancelFrame` is pushed through the pipeline and must reach the
  end. If it does not reach the end within the specified time, a warning is
  shown and the wait is aborted.

- Added a new "universal" (LLM-agnostic) `LLMContext` and accompanying
  `LLMContextAggregatorPair`, which will eventually replace `OpenAILLMContext`
  (and the other under-the-hood contexts) and the other context aggregators.
  The new universal `LLMContext` machinery allows a single context to be shared
  between different LLMs, enabling runtime LLM switching and scenarios like
  failover.

  From the developer's point of view, switching to using the new universal
  context machinery will usually be a matter of going from this:

  ```python
  context = OpenAILLMContext(messages, tools)
  context_aggregator = llm.create_context_aggregator(context)
  ```

  To this:

  ```python
  context = LLMContext(messages, tools)
  context_aggregator = LLMContextAggregatorPair(context)
  ```

  To start, the universal `LLMContext` is supported with the following LLM
  services:

  - `OpenAILLMService`
  - `GoogleLLMService`

- Added a new `LLMSwitcher` class to enable runtime LLM switching, built atop a
  new generic `ServiceSwitcher`.

  Switchers take a switching strategy. The first available strategy is
  `ServiceSwitcherStrategyManual`.

  To switch LLMs at runtime, the LLMs must be sharing one instance of the new
  universal `LLMContext` (see above bullet).

  ```python
  # Instantiate your LLM services
  llm_openai = OpenAILLMService(api_key=os.getenv("OPENAI_API_KEY"))
  llm_google = GoogleLLMService(api_key=os.getenv("GOOGLE_API_KEY"))

  # Instantiate a switcher
  # (ServiceSwitcherStrategyManual defaults to OpenAI, as it's first in the list)
  llm_switcher = LLMSwitcher(
      llms=[llm_openai, llm_google], strategy_type=ServiceSwitcherStrategyManual
  )

  # Create your pipeline
  pipeline = Pipeline(
      [
          transport.input(),
          stt,
          context_aggregator.user(),
          llm_switcher,
          tts,
          transport.output(),
          context_aggregator.assistant(),
      ]
  )
  task = PipelineTask(pipeline, params=PipelineParams(allow_interruptions=True))

  # ...
  # Whenever is appropriate, switch LLMs!
  await task.queue_frames([ManuallySwitchServiceFrame(service=llm_google)])
  ```

- Added an `LLMService.run_inference()` method to LLM services to enable
  direct, out-of-band (i.e. out-of-pipeline) inference.

### Changed

- Updated `daily-python` to 0.19.8.

- `PipelineTask` now waits for `StartFrame` to reach the end of the pipeline
  before pushing any other frames.

- Updated `CartesiaTTSService` and `CartesiaHttpTTSService` to align with
  Cartesia's changes for the `speed` parameter. It now takes only an enum of
  `slow`, `normal`, or `fast`.

- Added support to `AWSBedrockLLMService` for setting authentication
  credentials through environment variables.

- Updated `SarvamTTSService` to use WebSocket streaming for real-time audio
  generation with multiple Indian languages, with HTTP support still available
  via `SarvamHttpTTSService`.

### Fixed

- Fixed an RTVI issue that was causing frames to be pushed before pipeline was
  properly initialized.

- Fixed some `get_messages_for_logging()` that were returning a JSON string
  instead of a list.

- Fixed a `DailyTransport` issue that prevented DTMF tones from being sent.

- Fixed a missing import in `SentryMetrics`.

- Fixed `AWSPollyTTSService` to support AWS credential provider chain (IAM
  roles, IRSA, instance profiles) instead of requiring explicit environment
  variables.

- Fixed a `CartesiaTTSService` issue that was causing the application to hang
  after Cartesia's 5 minutes timed out.

- Fixed an issue preventing `SpeechmaticsSTTService` from transcribing audio.

## [0.0.81] - 2025-08-25

### Added

- Added `pipecat.extensions.voicemail`, a module for detecting voicemail vs.
  live conversation, primarily intended for use in outbound calling scenarios.
  The voicemail module is optimized for text LLMs only.

- Added new frames to the `idle_timeout_frames` arg: `TranscriptionFrame`,
  `InterimTranscriptionFrame`, `UserStartedSpeakingFrame`, and
  `UserStoppedSpeakingFrame`. These additions serve as indicators of user
  activity in the pipeline idle detection logic.

- Allow passing custom pipeline sink and source processors to a
  `Pipeline`. Pipeline source and sink processors are used to know and control
  what's coming in and out of a `Pipeline` processor.

- Added `FrameProcessor.pause_processing_system_frames()` and
  `FrameProcessor.resume_processing_system_frames()`. These allow to pause and
  resume the processing of system frame.

- Added new `on_process_frame()` observer method which makes it possible to know
  when a frame is being processed.

- Added new `FrameProcessor.entry_processor()` method. This allows you to access
  the first non-compound processor in a pipeline.

- Added `FrameProcessor` properties `processors`, `next` and `previous`.

- `ElevenLabsTTSService` now supports additional runtime changes to the `model`,
  `language`, and `voice_settings` parameters.

- Added `apply_text_normalization` support to `ElevenLabsTTSService` and
  `ElevenLabsHttpTTSService`.

- Added `MistralLLMService`, using Mistral's chat completion API.

- Added the ability to retry executing a chat completion after a timeout period
  for `OpenAILLMService` and its subclasses, `AnthropicLLMService`, and
  `AWSBedrockLLMService`. The LLM services accept new args:
  `retry_timeout_secs` and `retry_on_timeout`. This feature is disabled by
  default.

### Changed

- Updated `daily-python` to 0.19.7.

### Deprecated

- `FrameProcessor.wait_for_task()` is deprecated. Use `await task` or
  `await asyncio.wait_for(task, timeout)` instead.

### Removed

- Watchdog timers have been removed. They were introduced in 0.0.72 to help
  diagnose pipeline freezes. Unfortunately, they proved ineffective since they
  required developers to use Pipecat-specific queues, iterators, and events to
  correctly reset the timer, which limited their usefulness and added friction.

- Removed unused `FrameProcessor.set_parent()` and
  `FrameProcessor.get_parent()`.

### Fixed

- Fixed an issue that would cause `PipelineRunner` and `PipelineTask` to not
  handle external asyncio task cancellation properly.

- Added `SpeechmaticsSTTService` exception handling on connection and sending.

- Replaced `asyncio.wait_for()` for `wait_for2.wait_for()` for Python <
  3.12. because of issues regarding task cancellation (i.e. cancellation is
  never propagated).
  See https://bugs.python.org/issue42130

- Fixed an `AudioBufferProcessor` issues that would cause audio overlap when
  setting a max buffer size.

- Fixed an issue where `AsyncAITTSService` had very high latency in responding
  by adding `force=true` when sending the flush command.

### Performance

- Improve `PipelineTask` performance by using direct mode processors and by
  removing unnecessary tasks.

- Improve `ParallelPipeline` performance by using direct mode, by not
  creating a task for each frame and every sub-pipeline and also by removing
  other unnecessary tasks.

- `Pipeline` performance improvements by using direct mode.

### Other

- Added `14w-function-calling-mistal.py` using `MistralLLMService`.

- Added `13j-azure-transcription.py` using `AzureSTTService`.

## [0.0.80] - 2025-08-13

### Added

- Added `GeminiTTSService` which uses Google Gemini to generate TTS output. The
  Gemini model can be prompted to insert styled speech to control the TTS
  output.

- Added Exotel support to Pipecat's development runner. You can now connect
  using the runner with `uv run bot.py -t exotel` and an ngrok connection to
  HTTP port 7860.

- Added `enable_direct_mode` argument to `FrameProcessor`. The direct mode is
  for processors which require very little I/O or compute resources, that is
  processors that can perform their task almost immediately. These type of
  processors don't need any of the internal tasks and queues usually created by
  frame processors which means overall application performance might be slightly
  increased. Use with care.

- Added TTFB metrics for `HeyGenVideoService` and `TavusVideoService`.

- Added `endpoint_id` parameter to `AzureSTTService`. ([Custom EndpointId](https://docs.azure.cn/en-us/ai-services/speech-service/how-to-recognize-speech?pivots=programming-language-python#use-a-custom-endpoint))

### Changed

- `WatchdogPriorityQueue` now requires the items to be inserted to always be
  tuples and the size of the tuple needs to be specified in the constructor when
  creating the queue with the `tuple_size` argument.

- Updated Moondream to revision `2025-01-09`.

- Updated `PlayHTHttpTTSService` to no longer use the `pyht` client to remove
  compatibility issues with other packages. Now you can use the PlayHT HTTP
  service with other services, like GoogleLLMService.

- Updated `pyproject.toml` to once again pin `numba` to `==0.61.2` in order to
  resolve package versioning issues.

- Updated the `STTMuteFilter` to include `VADUserStartedSpeakingFrame` and
  `VADUserStoppedSpeakingFrame` in the list of frames to filter when the
  filtering is on.

### Performance

- Improving the latency of the `HeyGenVideoService`.

- Improved some frame processors performance by using the new frame processor
  direct mode. In direct mode a frame processor will process frames right away
  avoiding the need for internal queues and tasks. This is useful for some
  simple processors. For example, in processors that wrap other processors
  (e.g. `Pipeline`, `ParallelPipeline`), we add one processor before and one
  after the wrapped processors (internally, you will see them as sources and
  sinks). These sources and sinks don't do any special processing and they
  basically forward frames. So, for these simple processors we now enable the
  new direct mode which avoids creating any internal tasks (and queues) and
  therefore improves performance.

### Fixed

- Fixed an issue with the `BaseWhisperSTTService` where the language was
  specified as an enum and not a string.

- Fixed an issue where `SmallWebRTCTransport` ended before TTS finished.

- Fixed an issue in `OpenAIRealtimeBetaLLMService` where specifying a `text`
  `modalities` didn't result in text being outputted from the model.

- Added SSML reserved character escaping to `AzureBaseTTSService` to properly
  handle special characters in text sent to Azure TTS. This fixes an issue
  where characters like `&`, `<`, `>`, `"`, and `'` in LLM-generated text would
  cause TTS failures.

- Fixed a `WatchdogPriorityQueue` issue that could cause an exception when
  compating watchdog cancel sentinel items with other items in the queue.

- Fixed an issue that would cause system frames to not be processed with higher
  priority than other frames. This could cause slower interruption times.

- Fixed an issue where retrying a websocket connection error would result in an
  error.

### Other

- Add foundation example `19b-openai-realtime-beta-text.py`, showing how to use
  `OpenAIRealtimeBetaLLMService` to output text to a TTS service.

- Add vision support to release evals so we can run the foundational examples 12
  series.

- Added foundational example `15a-switch-languages.py` to release evals. It is
  able to detect if we switched the language properly.

- Updated foundational examples to show how to enclose complex logic
  (e.g. `ParallelPipeline`) into a single processor so the main pipeline becomes
  simpler.

- Added `07n-interruptible-gemini.py`, demonstrating how to use
  `GeminiTTSService`.

## [0.0.79] - 2025-08-07

### Changed

- Changed `pipecat-ai`'s `openai` dependency to `>=1.74.0,<=1.99.1` due to a
  breaking change in `openai` 1.99.2 ([commit](https://github.com/openai/openai-python/commit/657f551dbe583ffb259d987dafae12c6211fba06))

### Deprecated

- `TTSService.say()` is deprecated, push a `TTSSpeakFrame` instead. Calling
  functions directly is a discouraged pattern in Pipecat because, for example,
  it might cause issues with frame ordering.

- `LLMMessagesFrame` is deprecated, in favor of either:

  - `LLMMessagesUpdateFrame` with `run_llm=True`
  - `OpenAILLMContextFrame` with desired messages in a new context

- `LLMUserResponseAggregator` and `LLMAssistantResponseAggregator` are
  deprecated, as they depended on the now-deprecated `LLMMessagesFrame`. Use
  `LLMUserContextAggregator` and `LLMAssistantResponseAggregator` (or
  LLM-specific subclasses thereof) instead.

## [0.0.78] - 2025-08-07

### Added

- Added `SonioxSTTService` using Soniox's STT websocket API.

- Added `enable_emulated_vad_interruptions` to `LLMUserAggregatorParams`.
  When user speech is emulated (e.g. when a transcription is received but
  VAD doesn't detect speech), this parameter controls whether the emulated
  speech can interrupt the bot. Default is False (emulated speech is ignored
  while the bot is speaking).

- Added new `handle_sigint` and `handle_sigterm` to `RunnerArguments`. This
  allows applications to know what settings they should use for the environment
  they are running on. Also, added `pipeline_idle_timeout_secs` to be able to
  control the `PipelineTask` idle timeout.

- Added `processor` field to `ErrorFrame` to indicate `FrameProcessor` that
  generated the error.

- Added new language support for `AWSTranscribeSTTService`. All languages
  supporting streaming data input are now supported:
  https://docs.aws.amazon.com/transcribe/latest/dg/supported-languages.html

- Added support for Simli Trinity Avatars. A new `is_trinity_avatar` parameter
  has been introduced to specify whether the provided `faceId` corresponds to a
  Trinity avatar, which is required for optimal Trinity avatar performance.

- The development runner how handles custom `body` data for `DailyTransport`.
  The `body` data is passed to the Pipecat client. You can POST to the `/start`
  endpoint with a request body of:

  ```
  {
      "createDailyRoom": true,
      "dailyRoomProperties": { "start_video_off": true },
      "body": { "custom_data": "value" }
  }
  ```

  The `body` information is parsed and used in the application. The
  `dailyRoomProperties` are currently not handled.

- Added detailed latency logging to `UserBotLatencyLogObserver`, capturing
  average response time between user stop and bot start, as well as minimum and
  maximum response latency.

- Added Chinese, Japanese, Korean word timestamp support to
  `CartesiaTTSService`.

- Added `region` parameter to `GladiaSTTService`. Accepted values: eu-west
  (default), us-west.

### Changed

- System frames are now queued. Before, system frames could be generated from
  any task and would not guarantee any order which was causing undesired
  behavior. Also, it was possible to get into some rare recursion issues because
  of the way system frames were executed (they were executed in-place, meaning
  calling `push_frame()` would finish after the system frame traversed all the
  pipeline). This makes system frames more deterministic.

- Changed the default model for both `ElevenLabsTTSService` and
  `ElevenLabsHttpTTSService` to `eleven_turbo_v2_5`. The rationale for this
  change is that the Turbo v2.5 model exhibits the most stable voice quality
  along with very low latency TTFB; latencies are on par with the Flash v2.5
  model. Also, the Turbo v2.5 model outputs word/timestamp alignment data with
  correct spacing.

- The development runners `/connect` and `/start` endpoint now both return
  `dailyRoom` and `dailyToken` in place of the previous `room_url` and `token`.

- Updated the `pipecat.runner.daily` utility to only a take `DAILY_API_URL` and
  `DAILY_SAMPLE_ROOM_URL` environment variables instead of argparsing `-u` and
  `-k`, respectively.

- Updated `daily-python` to 0.19.6.

- Changed `TavusVideoService` to send audio or video frames only after the
  transport is ready, preventing warning messages at startup.

- The development runner now strips any provided protocol (e.g. https://) from
  the proxy address and issues a warning. It also strips trailing `/`.

### Deprecated

- In the `pipecat.runner.daily`, the `configure_with_args()` function is
  deprecated. Use the `configure()` function instead.

- The development runner's `/connect` endpoint is deprecated and will be
  removed in a future version. Use the `/start` endpoint in its place. In the
  meantime, both endpoints work and deliver equivalent functionality.

### Fixed

- Fixed a `DailyTransport` issue that would result in an unhandled
  `concurrent.futures.CancelledError` when a future is cancelled.

- Fixed a `RivaSTTService` issue that would result in an unhandled
  `concurrent.futures.CancelledError` when a future is cancelled when reading
  from the audio chunks from the incoming audio stream.

- Fixed an issue in the `BaseOutputTransport`, mainly reproducible with
  `FastAPIWebsocketOutputTransport` when the audio mixer was enabled, where the
  loop could consume 100% CPU by continuously returning without delay, preventing
  other asyncio tasks (such as cancellation or shutdown signals) from being
  processed.

- Fixed an issue where `BotStartedSpeakingFrame` and `BotStoppedSpeakingFrame`
  were not emitted when using `TavusVideoService` or `HeyGenVideoService`.

- Fixed an issue in `LiveKitTransport` where empty `AudioRawFrame`s were pushed
  down the pipeline. This resulted in warnings by the STT processor.
- Fixed `PiperTTSService` to send text as a JSON object in the request body,
  resolving compatibility with Piper's HTTP API.

- Fixed an issue with the `TavusVideoService` where an error was thrown due to
  missing transcription callbacks.

- Fixed an issue in `SpeechmaticsSTTService` where the `user_id` was set to
  `None` when diarization is not enabled.

### Performance

- Fixed an issue in `TaskObserver` (a proxy to all observers) that was degrading
  global performance.

### Other

- Added `07aa-interruptible-soniox.py`, `07ab-interruptible-inworld-http.py`,
  `07ac-interruptible-asyncai.py` and `07ac-interruptible-asyncai-http.py`
  release evals.

## [0.0.77] - 2025-07-31

### Added

- Added `InputTextRawFrame` frame type to handle user text input with Gemini
  Multimodal Live.

- Added `HeyGenVideoService`. This is an integration for HeyGen Interactive
  Avatar. A video service that handles audio streaming and requests HeyGen to
  generate avatar video responses. (see https://www.heygen.com/)

- Added the ability to switch voices to `RimeTTSService`.

- Added unified development runner for building voice AI bots across multiple
  transports

  - `pipecat.runner.run` – FastAPI-based development server with automatic bot
    discovery
  - `pipecat.runner.types` – Runner session argument types
    (`DailyRunnerArguments`, `SmallWebRTCRunnerArguments`,
    `WebSocketRunnerArguments`)
  - `pipecat.runner.utils.create_transport()` – Factory function for creating
    transports from session arguments
  - `pipecat.runner.daily` and `pipecat.runner.livekit` – Configuration
    utilities for Daily and LiveKit setups
  - Support for all transport types: Daily, WebRTC, Twilio, Telnyx, Plivo
  - Automatic telephony provider detection and serializer configuration
  - ESP32 WebRTC compatibility with SDP munging
  - Environment detection (`ENV=local`) for conditional features

- Added Async.ai TTS integration (https://async.ai/)

  - `AsyncAITTSService` – WebSocket-based streaming TTS with interruption
    support
  - `AsyncAIHttpTTSService` – HTTP-based streaming TTS service
  - Example scripts:
    - `examples/foundational/07ac-interruptible-asyncai.py` (WebSocket demo)
    - `examples/foundational/07ac-interruptible-asyncai-http.py` (HTTP demo)

- Added `transcription_bucket` params support to the `DailyRESTHelper`.

- Added a new TTS service, `InworldTTSService`. This service provides
  low-latency, high-quality speech generation using Inworld's streaming API.

- Added a new field `handle_sigterm` to `PipelineRunner`. It defaults to
  `False`. This field handles SIGTERM signals. The `handle_sigint` field still
  defaults to `True`, but now it handles only SIGINT signals.

- Added foundational example `14u-function-calling-ollama.py` for Ollama
  function calling.

- Added `LocalSmartTurnAnalyzerV2`, which supports local on-device inference
  with the new `smart-turn-v2` turn detection model.

- Added `set_log_level` to `DailyTransport`, allowing setting the logging level
  for Daily's internal logging system.

- Added `on_transcription_stopped` and `on_transcription_error` to Daily
  callbacks.

### Changed

- Changed the default `url` for `NeuphonicTTSService` to
  `wss://api.neuphonic.com` as it provides better global performance. You can
  set the URL to other URLs, such as the previous default:
  `wss://eu-west-1.api.neuphonic.com`.

- Update `daily-python` to 0.19.5.

- `STTMuteFilter` now pushes the `STTMuteFrame` upstream and downstream, to
  allow for more flexible `STTMuteFilter` placement.

- Play delayed messages from `ElevenLabsTTSService` if they still belong to the
  current context.

- Dependency compatibility improvements: Relaxed version constraints for core
  dependencies to support broader version ranges while maintaining stability:

  - `aiohttp`, `Markdown`, `nltk`, `numpy`, `Pillow`, `pydantic`, `openai`,
    `numba`: Now support up to the next major version (e.g. `numpy>=1.26.4,<3`)
  - `pyht`: Relaxed to `>=0.1.6` to resolve `grpcio` conflicts with
    `nvidia-riva-client`
  - `fastapi`: Updated to support versions `>=0.115.6,<0.117.0`
  - `torch`/`torchaudio`: Changed from exact pinning (`==2.5.0`) to compatible
    range (`~=2.5.0`)
  - `aws_sdk_bedrock_runtime`: Added Python 3.12+ constraint via environment
    marker
  - `numba`: Reduced minimum version to `0.60.0` for better compatibility

- Changed `NeuphonicHttpTTSService` to use a POST based request instead of the
  `pyneuphonic` package. This removes a package requirement, allowing Neuphonic
  to work with more services.

- Updated `ElevenLabsTTSService` to handle the case where
  `allow_interruptions=False`. Now, when interruptions are disabled, the same
  context ID will be used throughout the conversation.

- Updated the `deepgram` optional dependency to 4.7.0, which downgrades the
  `tasks cancelled error` to a debug log. This removes the log from appearing
  in Pipecat logs upon leaving.

- Upgraded the `websockets` implementation to the new asyncio implementation.
  Along with this change, we're updating support for versions >=13.1.0 and
  <15.0.0. All services have been update to use the asyncio implementation.

- Updated `MiniMaxHttpTTSService` with a `base_url` arg where you can specify
  the Global endpoint (default) or Mainland China.

- Replaced regex-based sentence detection in `match_endofsentence` with NLTK's
  punkt_tab tokenizer for more reliable sentence boundary detection.

- Changed the `livekit` optional dependency for `tenacity` to
  `tenacity>=8.2.3,<10.0.0` in order to support the `google-genai` package.

- For `LmntTTSService`, changed the default `model` to `blizzard`, LMNT's
  recommended model.

- Updated `SpeechmaticsSTTService`:
  - Added support for additional diarization options.
  - Added foundational example `07a-interruptible-speechmatics-vad.py`, which
    uses VAD detection provided by `SpeechmaticsSTTService`.

### Fixed

- Fixed a `LLMUserResponseAggregator` issue where interruptions were not being
  handled properly.

- Fixed `PiperTTSService` to work with newer Piper GPL.

- Fixed a race condition in `FastAPIWebsocketClient` that occurred when
  attempting to send a message while the client was disconnecting.

- Fixed an issue in `GoogleLLMService` where interruptions did not work when an
  interruption strategy was used.

- Fixed an issue in the `TranscriptProcessor` where newline characters could
  cause the transcript output to be corrupted (e.g. missing all spaces).

- Fixed an issue in `AudioBufferProcessor` when using `SmallWebRTCTransport`
  where, if the microphone was muted, track timing was not respected.

- Fixed an error that occurs when pushing an `LLMMessagesFrame`. Only some LLM
  services, like Grok, are impacted by this issue. The fix is to remove the
  optional `name` property that was being added to the message.

- Fixed an issue in `AudioBufferProcessor` that caused garbled audio when
  `enable_turn_audio` was enabled and audio resampling was required.

- Fixed a dependency issue for uv users where an `llvmlite` version required
  python 3.9.

- Fixed an issue in `MiniMaxHttpTTSService` where the `pitch` param was the
  incorrect type.

- Fixed an issue with OpenTelemetry tracing where the `enable_tracing` flag did
  not disable the internal tracing decorator functions.

- Fixed an issue in `OLLamaLLMService` where kwargs were not passed correctly
  to the parent class.

- Fixed an issue in `ElevenLabsTTSService` where the word/timestamp pairs were
  calculating word boundaries incorrectly.

- Fixed an issue where, in some edge cases, the
  `EmulateUserStartedSpeakingFrame` could be created even if we didn't have a
  transcription.

- Fixed an issue in `GoogleLLMContext` where it would inject the
  `system_message` as a "user" message into cases where it was not meant to;
  it was only meant to do that when there were no "regular" (non-function-call)
  messages in the context, to ensure that inference would run properly.

- Fixed an issue in `LiveKitTransport` where the `on_audio_track_subscribed` was
  never emitted.

### Other

- Added new quickstart demos:

  - examples/quickstart: voice AI bot quickstart
  - examples/client-server-web: client/server starter example
  - examples/phone-bot-twilio: twilio starter example

- Removed most of the examples from the pipecat repo. Examples can now be
  found in: https://github.com/pipecat-ai/pipecat-examples.

## [0.0.76] - 2025-07-11

### Added

- Added `SpeechControlParamsFrame`, a new `SystemFrame` that notifies
  downstream processors of the VAD and Turn analyzer params. This frame is
  pushed by the `BaseInputTransport` at Start and any time a
  `VADParamsUpdateFrame` is received.

### Changed

- Two package dependencies have been updated:
  - `numpy` now supports 1.26.0 and newer
  - `transformers` now supports 4.48.0 and newer

### Fixed

- Fixed an issue with RTVI's handling of `append-to-context`.

- Fixed an issue where using audio input with a sample rate requiring resampling
  could result in empty audio being passed to STT services, causing errors.

- Fixed the VAD analyzer to process the full audio buffer as long as it contains
  more than the minimum required bytes per iteration, instead of only analyzing
  the first chunk.

- Fixed an issue in ParallelPipeline that caused errors when attempting to drain
  the queues.

- Fixed an issue with emulated VAD timeout inconsistency in
  `LLMUserContextAggregator`. Previously, emulated VAD scenarios (where
  transcription is received without VAD detection) used a hardcoded
  `aggregation_timeout` (default 0.5s) instead of matching the VAD's
  `stop_secs` parameter (default 0.8s). This created different user experiences
  between real VAD and emulated VAD scenarios. Now, emulated VAD timeouts
  automatically synchronize with the VAD's `stop_secs` parameter.

- Fix a pipeline freeze when using AWS Nova Sonic, which would occur if the
  user started early, while the bot was still working through
  `trigger_assistant_response()`.

## [0.0.75] - 2025-07-08 [YANKED]

**This release has been yanked due to resampling issues affecting audio output
quality and critical bugs impacting `ParallelPipelines` functionality.**

**Please upgrade to version 0.0.76 or later.**

### Added

- Added an `aggregate_sentences` arg in `CartesiaTTSService`,
  `ElevenLabsTTSService`, `NeuphonicTTSService` and `RimeTTSService`, where the
  default value is True. When `aggregate_sentences` is True, the `TTSService`
  aggregates the LLM streamed tokens into sentences by default. Note: setting
  the value to False requires a custom processor before the `TTSService` to
  aggregate LLM tokens.

- Added `kwargs` to the `OLLamaLLMService` to allow for configuration args to
  be passed to Ollama.

- Added call hang-up error handling in `TwilioFrameSerializer`, which handles
  the case where the user has hung up before the `TwilioFrameSerializer` hangs
  up the call.

### Changed

- Updated `RTVIObserver` and `RTVIProcessor` to match the new RTVI 1.0.0 protocol.
  This includes:

  - Deprecating support for all messages related to service configuaration and
    actions.
  - Adding support for obtaining and logging data about client, including its
    RTVI version and optionally included system information (OS/browser/etc.)
  - Adding support for handling the new `client-message` RTVI message through
    either a `on_client_message` event handler or listening for a new
    `RTVIClientMessageFrame`
  - Adding support for responding to a `client-message` with a `server-response`
    via either a direct call on the `RTVIProcessor` or via pushing a new
    `RTVIServerResponseFrame`
  - Adding built-in support for handling the new `append-to-context` RTVI message
    which allows a client to add to the user or assistant llm context. No extra
    code is required for supporting this behavior.
  - Updating all JavaScript and React client RTVI examples to use versions 1.0.0
    of the clients.

  Get started migrating to RTVI protocol 1.0.0 by following the migration guide:
  https://docs.pipecat.ai/client/migration-guide

- Refactored `AWSBedrockLLMService` and `AWSPollyTTSService` to work
  asynchronously using `aioboto3` instead of the `boto3` library.

- The `UserIdleProcessor` now handles the scenario where function calls take
  longer than the idle timeout duration. This allows you to use the
  `UserIdleProcessor` in conjunction with function calls that take a while to
  return a result.

### Fixed

- Updated the `NeuphonicTTSService` to work with the updated websocket API.

- Fixed an issue with `RivaSTTService` where the watchdog feature was causing
  an error on initialization.

### Performance

- Remove unncessary push task in each `FrameProcessor`.

## [0.0.74] - 2025-07-03 [YANKED]

**This release has been yanked due to resampling issues affecting audio output
quality and critical bugs impacting `ParallelPipelines` functionality.**

**Please upgrade to version 0.0.76 or later.**

### Added

- Added a new STT service, `SpeechmaticsSTTService`. This service provides
  real-time speech-to-text transcription using the Speechmatics API. It supports
  partial and final transcriptions, multiple languages, various audio formats,
  and speaker diarization.

- Added `normalize` and `model_id` to `FishAudioTTSService`.

- Added `http_options` argument to `GoogleLLMService`.

- Added `run_llm` field to `LLMMessagesAppendFrame` and `LLMMessagesUpdateFrame`
  frames. If true, a context frame will be pushed triggering the LLM to respond.

- Added a new `SOXRStreamAudioResampler` for processing audio in chunks or
  streams. If you write your own processor and need to use an audio resampler,
  use the new `create_stream_resampler()`.

- Added new `DailyParams.audio_in_user_tracks` to allow receiving one track per
  user (default) or a single track from the room (all participants mixed).

- Added support for providing "direct" functions, which don't need an
  accompanying `FunctionSchema` or function definition dict. Instead, metadata
  (i.e. `name`, `description`, `properties`, and `required`) are automatically
  extracted from a combination of the function signature and docstring.

  Usage:

  ```python
  # "Direct" function
  # `params` must be the first parameter
  async def do_something(params: FunctionCallParams, foo: int, bar: str = ""):
      """
      Do something interesting.

      Args:
        foo (int): The foo to do something interesting with.
        bar (string): The bar to do something interesting with.
      """

      result = await process(foo, bar)
      await params.result_callback({"result": result})


  # ...

  llm.register_direct_function(do_something)

  # ...

  tools = ToolsSchema(standard_tools=[do_something])
  ```

- `user_id` is now populated in the `TranscriptionFrame` and
  `InterimTranscriptionFrame` when using a transport that provides a `user_id`,
  like `DailyTransport` or `LiveKitTransport`.

- Added `watchdog_coroutine()`. This is a watchdog helper for couroutines. So,
  if you have a coroutine that is waiting for a result and that takes a long
  time, you will need to wrap it with `watchdog_coroutine()` so the watchdog
  timers are reset regularly.

- Added `session_token` parameter to `AWSNovaSonicLLMService`.

- Added Gemini Multimodal Live File API for uploading, fetching, listing, and
  deleting files. See `26f-gemini-live-files-api.py` for example usage.

### Changed

- Updated all the services to use the new `SOXRStreamAudioResampler`, ensuring smooth
  transitions and eliminating clicks.

- Upgraded `daily-python` to 0.19.4.

- Updated `google` optional dependency to use `google-genai` version `1.24.0`.

### Fixed

- Fixed an issue where audio would get stuck in the queue when an interrupt occurs
  during Azure TTS synthesis.

- Fixed a race condition that occurs in Python 3.10+ where the task could miss
  the `CancelledError` and continue running indefinitely, freezing the pipeline.

- Fixed a `AWSNovaSonicLLMService` issue introduced in 0.0.72.

### Deprecated

- In `FishAudioTTSService`, deprecated `model` and replaced with
  `reference_id`. This change is to better align with Fish Audio's variable
  naming and to reduce confusion about what functionality the variable
  controls.

## [0.0.73] - 2025-06-26

### Fixed

- Fixed an issue introduced in 0.0.72 that would cause `ElevenLabsTTSService`,
  `GladiaSTTService`, `NeuphonicTTSService` and `OpenAIRealtimeBetaLLMService`
  to throw an error.

## [0.0.72] - 2025-06-26

### Added

- Added logging and improved error handling to help diagnose and prevent potential
  Pipeline freezes.

- Added `WatchdogQueue`, `WatchdogPriorityQueue`, `WatchdogEvent` and
  `WatchdogAsyncIterator`. These helper utilities reset watchdog timers
  appropriately before they expire. When watchdog timers are disabled, the
  utilities behave as standard counterparts without side effects.

- Introduce task watchdog timers. Watchdog timers are used to detect if a
  Pipecat task is taking longer than expected (by default 5 seconds). Watchdog
  timers are disabled by default and can be enabled globally by passing
  `enable_watchdog_timers` argument to `PipelineTask` constructor. It is
  possible to change the default watchdog timer timeout by using the
  `watchdog_timeout` argument. You can also log how long it takes to reset the
  watchdog timers which is done with the `enable_watchdog_logging`. You can
  control all these settings per each frame processor or even per task. That is,
  you can set `enable_watchdog_timers`, `enable_watchdog_logging` and
  `watchdog_timeout` when creating any frame processor through their constructor
  arguments or when you create a task with `FrameProcessor.create_task()`. Note
  that watchdog timers only work with Pipecat tasks and will not work if you use
  `asycio.create_task()` or similar.

- Added `lexicon_names` parameter to `AWSPollyTTSService.InputParams`.

- Added reconnection logic and audio buffer management to `GladiaSTTService`.

- The `TurnTrackingObserver` now ends a turn upon observing an `EndFrame` or
  `CancelFrame`.

- Added Polish support to `AWSTranscribeSTTService`.

- Added new frames `FrameProcessorPauseFrame` and `FrameProcessorResumeFrame`
  which allow pausing and resuming frame processing for a given frame
  processor. These are control frames, so they are ordered. Pausing frame
  processor will keep old frames in the internal queues until resume takes
  place. Frames being pushed while a frame processor is paused will be pushed to
  the queues. When frame processing is resumed all queued frames will be
  processed in order. Also added `FrameProcessorPauseUrgentFrame` and
  `FrameProcessorResumeUrgentFrame` which are system frames and therefore they
  have high priority.

- Added a property called `has_function_calls_in_progress` in
  `LLMAssistantContextAggregator` that exposes whether a function call is in
  progress.

- Added `SambaNovaLLMService` which provides llm api integration with an
  OpenAI-compatible interface.

- Added `SambaNovaTTSService` which provides speech-to-text functionality using
  SambaNovas's (whisper) API.

- Add fundational examples for function calling and transcription
  `14s-function-calling-sambanova.py`, `13g-sambanova-transcription.py`

### Changed

- `HeartbeatFrame`s are now control frames. This will make it easier to detect
  pipeline freezes. Previously, heartbeat frames were system frames which meant
  they were not get queued with other frames, making it difficult to detect
  pipeline stalls.

- Updated `OpenAIRealtimeBetaLLMService` to accept `language` in the
  `InputAudioTranscription` class for all models.

- Updated the default model for `OpenAIRealtimeBetaLLMService` to
  `gpt-4o-realtime-preview-2025-06-03`.

- The `PipelineParams` arg `allow_interruptions` now defaults to `True`.

- `TavusTransport` and `TavusVideoService` now send audio to Tavus using WebRTC
  audio tracks instead of `app-messages` over WebSocket. This should improve the
  overall audio quality.

- Upgraded `daily-python` to 0.19.3.

### Fixed

- Fixed an issue that would cause heartbeat frames to be sent before processors
  were started.

- Fixed an event loop blocking issue when using `SentryMetrics`.

- Fixed an issue in `FastAPIWebsocketClient` to ensure proper disconnection
  when the websocket is already closed.

- Fixed an issue where the `UserStoppedSpeakingFrame` was not received if the
  transport was not receiving new audio frames.

- Fixed an edge case where if the user interrupted the bot but no new aggregation
  was received, the bot would not resume speaking.

- Fixed an issue with `TelnyxFrameSerializer` where it would throw an exception
  when the user hung up the call.

- Fixed an issue with `ElevenLabsTTSService` where the context was not being
  closed.

- Fixed function calling in `AWSNovaSonicLLMService`.

- Fixed an issue that would cause multiple `PipelineTask.on_idle_timeout`
  events to be triggered repeatedly.

- Fixed an issue that was causing user and bot speech to not be synchronized
  during recordings.

- Fixed an issue where voice settings weren't applied to ElevenLabsTTSService.

- Fixed an issue with `GroqTTSService` where it was not properly parsing the
  WAV file header.

- Fixed an issue with `GoogleSTTService` where it was constantly reconnecting
  before starting to receive audio from the user.

- Fixed an issue where `GoogleLLMService`'s TTFB value was incorrect.

### Deprecated

- `AudioBufferProcessor` parameter `user_continuos_stream` is deprecated.

### Other

- Rename `14e-function-calling-gemini.py` to `14e-function-calling-google.py`.

## [0.0.71] - 2025-06-10

### Added

- Adds a parameter called `additional_span_attributes` to PipelineTask that
  lets you add any additional attributes you'd like to the conversation span.

### Fixed

- Fixed an issue with `CartesiaSTTService` initialization.

## [0.0.70] - 2025-06-10

### Added

- Added `ExotelFrameSerializer` to handle telephony calls via Exotel.

- Added the option `informal` to `TranslationConfig` on Gladia config.
  Allowing to force informal language forms when available.

- Added `CartesiaSTTService` which is a websocket based implementation to
  transcribe audio. Added a foundational example in
  `13f-cartesia-transcription.py`

- Added an `websocket` example, showing how to use the new Pipecat client
  `WebsocketTransport` to connect with Pipecat `FastAPIWebsocketTransport` or
  `WebsocketServerTransport`.

- Added language support to `RimeHttpTTSService`. Extended languages to include
  German and French for both `RimeTTSService` and `RimeHttpTTSService`.

### Changed

- Upgraded `daily-python` to 0.19.2.

- Make `PipelineTask.add_observer()` synchronous. This allows callers to call it
  before doing the work of running the `PipelineTask` (i.e. without invoking
  `PipelineTask.set_event_loop()` first).

- Pipecat 0.0.69 forced `uvloop` event loop on Linux on macOS. Unfortunately,
  this is causing issue in some systems. So, `uvloop` is not enabled by default
  anymore. If you want to use `uvloop` you can just set the `asyncio` event
  policy before starting your agent with:

```python
asyncio.set_event_loop_policy(uvloop.EventLoopPolicy())
```

### Fixed

- Fixed an issue with various TTS services that would cause audio glitches at
  the start of every bot turn.

- Fixed an `ElevenLabsTTSService` issue where a context warning was printed
  when pushing a `TTSSpeakFrame`.

- Fixed an `AssemblyAISTTService` issue that could cause unexpected behavior
  when yielding empty `Frame()`s.

- Fixed an issue where `OutputAudioRawFrame.transport_destination` was being
  reset to `None` instead of retaining its intended value before sending the
  audio frame to `write_audio_frame`.

- Fixed a typo in Livekit transport that prevented initialization.

## [0.0.69] - 2025-06-02 "AI Engineer World's Fair release" ✨

### Added

- Added a new frame `FunctionCallsStartedFrame`. This frame is pushed both
  upstream and downstream from the LLM service to indicate that one or more
  function calls are going to be executed.

- Added LLM services `on_function_calls_started` event. This event will be
  triggered when the LLM service receives function calls from the model and is
  going to start executing them.

- Function calls can now be executed sequentially (in the order received in the
  completion) by passing `run_in_parallel=False` when creating your LLM
  service. By default, if the LLM completion returns 2 or more function calls
  they run concurrently. In both cases, concurrently and sequentially, a new LLM
  completion will run when the last function call finishes.

- Added OpenTelemetry tracing for `GeminiMultimodalLiveLLMService` and
  `OpenAIRealtimeBetaLLMService`.

- Added initial support for interruption strategies, which determine if the user
  should interrupt the bot while the bot is speaking. Interruption strategies
  can be based on factors such as audio volume or the number of words spoken by
  the user. These can be specified via the new `interruption_strategies` field
  in `PipelineParams`. A new `MinWordsInterruptionStrategy` strategy has been
  introduced which triggers an interruption if the user has spoken a minimum
  number of words. If no interruption strategies are specified, the normal
  interruption behavior applies. If multiple strategies are provided, the first
  one that evaluates to true will trigger the interruption.

- `BaseInputTransport` now handles `StopFrame`. When a `StopFrame` is received
  the transport will pause sending frames downstream until a new `StartFrame` is
  received. This allows the transport to be reused (keeping the same connection)
  in a different pipeline.

- Updated AssemblyAI STT service to support their latest streaming
  speech-to-text model with improved transcription latency and endpointing.

- You can now access STT service results through the new
  `TranscriptionFrame.result` and `InterimTranscriptionFrame.result` field. This
  is useful in case you use some specific settings for the STT and you want to
  access the STT results.

- The examples runner is now public from the `pipecat.examples` package. This
  allows everyone to build their own examples and run them easily.

- It is now possible to push `OutputDTMFFrame` or `OutputDTMFUrgentFrame` with
  `DailyTransport`. This will be sent properly if a Daily dial-out connection
  has been established.

- Added `OutputDTMFUrgentFrame` to send a DTMF keypress quickly. The previous
  `OutputDTMFFrame` queues the keypress with the rest of data frames.

- Added `DTMFAggregator`, which aggregates keypad presses into
  `TranscriptionFrame`s. Aggregation occurs after a timeout, termination key
  press, or user interruption. You can specify the prefix of the
  `TranscriptionFrame`.

- Added new functions `DailyTransport.start_transcription()` and
  `DailyTransport.stop_transcription()` to be able to start and stop Daily
  transcription dynamically (maybe with different settings).

### Changed

- Reverted the default model for `GeminiMultimodalLiveLLMService` back to
  `models/gemini-2.0-flash-live-001`.
  `gemini-2.5-flash-preview-native-audio-dialog` has inconsistent performance.
  You can opt in to using this model by setting the `model` arg.

- Function calls are now cancelled by default if there's an interruption. To
  disable this behavior you can set `cancel_on_interruption=False` when
  registering the function call. Since function calls are executed as tasks you
  can tell if a function call has been cancelled by catching the
  `asyncio.CancelledError` exception (and don't forget to raise it again!).

- Updated OpenTelemetry tracing attribute `metrics.ttfb_ms` to `metrics.ttfb`.
  The attribute reports TTFB in seconds.

### Deprecated

- `DailyTransport.send_dtmf()` is deprecated, push an `OutputDTMFFrame` or an
  `OutputDTMFUrgentFrame` instead.

### Fixed

- Fixed an issue with `ElevenLabsTTSService` where long responses would
  continue generating output even after an interruption.

- Fixed an issue with the `OpenAILLMContext` where non-Roman characters were
  being incorrectly encoded as Unicode escape sequences. This was a logging
  issue and did not impact the actual conversation.

- In `AWSBedrockLLMService`, worked around a possible bug in AWS Bedrock where
  a `toolConfig` is required if there has been previous tool use in the
  messages array. This workaround includes a no_op factory function call is
  used to satisfy the requirement.

- Fixed `WebsocketClientTransport` to use `FrameProcessorSetup.task_manager`
  instead of `StartFrame.task_manager`.

### Performance

- Use `uvloop` as the new event loop on Linux and macOS systems.

## [0.0.68] - 2025-05-28

### Added

- Added `GoogleHttpTTSService` which uses Google's HTTP TTS API.

- Added `TavusTransport`, a new transport implementation compatible with any
  Pipecat pipeline. When using the `TavusTransport`the Pipecat bot will
  connect in the same room as the Tavus Avatar and the user.

- Added `PlivoFrameSerializer` to support Plivo calls. A full running example
  has also been added to `examples/plivo-chatbot`.

- Added `UserBotLatencyLogObserver`. This is an observer that logs the latency
  between when the user stops speaking and when the bot starts speaking. This
  gives you an initial idea on how quickly the AI services respond.

- Added `SarvamTTSService`, which implements Sarvam AI's TTS API:
  https://docs.sarvam.ai/api-reference-docs/text-to-speech/convert.

- Added `PipelineTask.add_observer()` and `PipelineTask.remove_observer()` to
  allow mangaging observers at runtime. This is useful for cases where the task
  is passed around to other code components that might want to observe the
  pipeline dynamically.

- Added `user_id` field to `TranscriptionMessage`. This allows identifying the
  user in a multi-user scenario. Note that this requires that
  `TranscriptionFrame` has the `user_id` properly set.

- Added new `PipelineTask` event handlers `on_pipeline_started`,
  `on_pipeline_stopped`, `on_pipeline_ended` and `on_pipeline_cancelled`, which
  correspond to the `StartFrame`, `StopFrame`, `EndFrame` and `CancelFrame`
  respectively.

- Added additional languages to `LmntTTSService`. Languages include: `hi`,
  `id`, `it`, `ja`, `nl`, `pl`, `ru`, `sv`, `th`, `tr`, `uk`, `vi`.

- Added a `model` parameter to the `LmntTTSService` constructor, allowing
  switching between LMNT models.

- Added `MiniMaxHttpTTSService`, which implements MiniMax's T2A API for TTS.
  Learn more: https://www.minimax.io/platform_overview

- A new function `FrameProcessor.setup()` has been added to allow setting up
  frame processors before receiving a `StartFrame`. This is what's happening
  internally: `FrameProcessor.setup()` is called, `StartFrame` is pushed from
  the beginning of the pipeline, your regular pipeline operations, `EndFrame`
  or `CancelFrame` are pushed from the beginning of the pipeline and finally
  `FrameProcessor.cleanup()` is called.

- Added support for OpenTelemetry tracing in Pipecat. This initial
  implementation includes:

  - A `setup_tracing` method where you can specify your OpenTelemetry exporter
  - Service decorators for STT (`@traced_stt`), LLM (`@traced_llm`), and TTS
    (`@traced_tts`) which trace the execution and collect properties and
    metrics (TTFB, token usage, character counts, etc.)
  - Class decorators that provide execution tracking; these are generic and can
    be used for service tracking as needed
  - Spans that help track traces on a per conversations and turn basis:

  ```
  conversation-uuid
  ├── turn-1
  │   ├── stt_deepgramsttservice
  │   ├── llm_openaillmservice
  │   └── tts_cartesiattsservice
  ...
  └── turn-n
      └── ...
  ```

  By default, Pipecat has implemented service decorators to trace execution of
  STT, LLM, and TTS services. You can enable tracing by setting
  `enable_tracing` to `True` in the PipelineTask.

- Added `TurnTrackingObserver`, which tracks the start and end of a user/bot
  turn pair and emits events `on_turn_started` and `on_turn_stopped`
  corresponding to the start and end of a turn, respectively.

- Allow passing observers to `run_test()` while running unit tests.

### Changed

- Upgraded `daily-python` to 0.19.1.

- ⚠️ Updated `SmallWebRTCTransport` to align with how other transports handle
  `on_client_disconnected`. Now, when the connection is closed and no reconnection
  is attempted, `on_client_disconnected` is called instead of `on_client_close`. The
  `on_client_close` callback is no longer used, use `on_client_disconnected` instead.

- Check if `PipelineTask` has already been cancelled.

- Don't raise an exception if event handler is not registered.

- Upgraded `deepgram-sdk` to 4.1.0.

- Updated `GoogleTTSService` to use Google's streaming TTS API. The default
  voice also updated to `en-US-Chirp3-HD-Charon`.

- ⚠️ Refactored the `TavusVideoService`, so it acts like a proxy, sending audio
  to Tavus and receiving both audio and video. This will make
  `TavusVideoService` usable with any Pipecat pipeline and with any transport.
  This is a **breaking change**, check the
  `examples/foundational/21a-tavus-layer-small-webrtc.py` to see how to use it.

- `DailyTransport` now uses custom microphone audio tracks instead of virtual
  microphones. Now, multiple Daily transports can be used in the same process.

- `DailyTransport` now captures audio from individual participants instead of
  the whole room. This allows identifying audio frames per participant.

- Updated the default model for `AnthropicLLMService` to
  `claude-sonnet-4-20250514`.

- Updated the default model for `GeminiMultimodalLiveLLMService` to
  `models/gemini-2.5-flash-preview-native-audio-dialog`.

- `BaseTextFilter` methods `filter()`, `update_settings()`,
  `handle_interruption()` and `reset_interruption()` are now async.

- `BaseTextAggregator` methods `aggregate()`, `handle_interruption()` and
  `reset()` are now async.

- The API version for `CartesiaTTSService` and `CartesiaHttpTTSService` has
  been updated. Also, the `cartesia` dependency has been updated to 2.x.

- `CartesiaTTSService` and `CartesiaHttpTTSService` now support Cartesia's new
  `speed` parameter which accepts values of `slow`, `normal`, and `fast`.

- `GeminiMultimodalLiveLLMService` now uses the user transcription and usage
  metrics provided by Gemini Live.

- `GoogleLLMService` has been updated to use `google-genai` instead of the
  deprecated `google-generativeai`.

### Deprecated

- In `CartesiaTTSService` and `CartesiaHttpTTSService`, `emotion` has been
  deprecated by Cartesia. Pipecat is following suit and deprecating `emotion`
  as well.

### Removed

- Since `GeminiMultimodalLiveLLMService` now transcribes it's own audio, the
  `transcribe_user_audio` arg has been removed. Audio is now transcribed
  automatically.

- Removed `SileroVAD` frame processor, just use `SileroVADAnalyzer`
  instead. Also removed, `07a-interruptible-vad.py` example.

### Fixed

- Fixed a `DailyTransport` issue that was not allow capturing video frames if
  framerate was greater than zero.

- Fixed a `DeegramSTTService` connection issue when the user provided their own
  `LiveOptions`.

- Fixed a `DailyTransport` issue that would cause images needing resize to block
  the event loop.

- Fixed an issue with `ElevenLabsTTSService` where changing the model or voice
  while the service is running wasn't working.

- Fixed an issue that would cause multiple instances of the same class to behave
  incorrectly if any of the given constructor arguments defaulted to a mutable
  value (e.g. lists, dictionaries, objects).

- Fixed an issue with `CartesiaTTSService` where `TTSTextFrame` messages weren't
  being emitted when the model was set to `sonic`. This resulted in the
  assistant context not being updated with assistant messages.

### Performance

- `DailyTransport`: process audio, video and events in separate tasks.

- Don't create event handler tasks if no user event handlers have been
  registered.

### Other

- It is now possible to run all (or most) foundational example with multiple
  transports. By default, they run with P2P (Peer-To-Peer) WebRTC so you can try
  everything locally. You can also run them with Daily or even with a Twilio
  phone number.

- Added foundation examples `07y-interruptible-minimax.py` and
  `07z-interruptible-sarvam.py`to show how to use the `MiniMaxHttpTTSService`
  and `SarvamTTSService`, respectively.

- Added an `open-telemetry-tracing` example, showing how to setup tracing. The
  example also includes Jaeger as an open source OpenTelemetry client to review
  traces from the example runs.

- Added foundational example `29-turn-tracking-observer.py` to show how to use
  the `TurnTrackingObserver`.

## [0.0.67] - 2025-05-07

### Added

- Added `DebugLogObserver` for detailed frame logging with configurable
  filtering by frame type and endpoint. This observer automatically extracts
  and formats all frame data fields for debug logging.

- `UserImageRequestFrame.video_source` field has been added to request an image
  from the desired video source.

- Added support for the AWS Nova Sonic speech-to-speech model with the new
  `AWSNovaSonicLLMService`.
  See https://docs.aws.amazon.com/nova/latest/userguide/speech.html.
  Note that it requires Python >= 3.12 and `pip install pipecat-ai[aws-nova-sonic]`.

- Added new AWS services `AWSBedrockLLMService` and `AWSTranscribeSTTService`.

- Added `on_active_speaker_changed` event handler to the `DailyTransport` class.

- Added `enable_ssml_parsing` and `enable_logging` to `InputParams` in
  `ElevenLabsTTSService`.

- Added support to `RimeHttpTTSService` for the `arcana` model.

### Changed

- Updated `ElevenLabsTTSService` to use the beta websocket API
  (multi-stream-input). This new API supports context_ids and cancelling those
  contexts, which greatly improves interruption handling.

- Observers `on_push_frame()` now take a single argument `FramePushed` instead
  of multiple arguments.

- Updated the default voice for `DeepgramTTSService` to `aura-2-helena-en`.

### Deprecated

- `PollyTTSService` is now deprecated, use `AWSPollyTTSService` instead.

- Observer `on_push_frame(src, dst, frame, direction, timestamp)` is now
  deprecated, use `on_push_frame(data: FramePushed)` instead.

### Fixed

- Fixed a `DailyTransport` issue that was causing issues when multiple audio or
  video sources where being captured.

- Fixed a `UltravoxSTTService` issue that would cause the service to generate
  all tokens as one word.

- Fixed a `PipelineTask` issue that would cause tasks to not be cancelled if
  task was cancelled from outside of Pipecat.

- Fixed a `TaskManager` that was causing dangling tasks to be reported.

- Fixed an issue that could cause data to be sent to the transports when they
  were still not ready.

- Remove custom audio tracks from `DailyTransport` before leaving.

### Removed

- Removed `CanonicalMetricsService` as it's no longer maintained.

## [0.0.66] - 2025-05-02

### Added

- Added two new input parameters to `RimeTTSService`: `pause_between_brackets`
  and `phonemize_between_brackets`.

- Added support for cross-platform local smart turn detection. You can use
  `LocalSmartTurnAnalyzer` for on-device inference using Torch.

- `BaseOutputTransport` now allows multiple destinations if the transport
  implementation supports it (e.g. Daily's custom tracks). With multiple
  destinations it is possible to send different audio or video tracks with a
  single transport simultaneously. To do that, you need to set the new
  `Frame.transport_destination` field with your desired transport destination
  (e.g. custom track name), tell the transport you want a new destination with
  `TransportParams.audio_out_destinations` or
  `TransportParams.video_out_destinations` and the transport should take care of
  the rest.

- Similar to the new `Frame.transport_destination`, there's a new
  `Frame.transport_source` field which is set by the `BaseInputTransport` if the
  incoming data comes from a non-default source (e.g. custom tracks).

- `TTSService` has a new `transport_destination` constructor parameter. This
  parameter will be used to update the `Frame.transport_destination` field for
  each generated `TTSAudioRawFrame`. This allows sending multiple bots' audio to
  multiple destinations in the same pipeline.

- Added `DailyTransportParams.camera_out_enabled` and
  `DailyTransportParams.microphone_out_enabled` which allows you to
  enable/disable the main output camera or microphone tracks. This is useful if
  you only want to use custom tracks and not send the main tracks. Note that you
  still need `audio_out_enabled=True` or `video_out_enabled`.

- Added `DailyTransport.capture_participant_audio()` which allows you to capture
  an audio source (e.g. "microphone", "screenAudio" or a custom track name) from
  a remote participant.

- Added `DailyTransport.update_publishing()` which allows you to update the call
  video and audio publishing settings (e.g. audio and video quality).

- Added `RTVIObserverParams` which allows you to configure what RTVI messages
  are sent to the clients.

- Added a `context_window_compression` InputParam to
  `GeminiMultimodalLiveLLMService` which allows you to enable a sliding context
  window for the session as well as set the token limit of the sliding window.

- Updated `SmallWebRTCConnection` to support `ice_servers` with credentials.

- Added `VADUserStartedSpeakingFrame` and `VADUserStoppedSpeakingFrame`,
  indicating when the VAD detected the user to start and stop speaking. These
  events are helpful when using smart turn detection, as the user's stop time
  can differ from when their turn ends (signified by UserStoppedSpeakingFrame).

- Added `TranslationFrame`, a new frame type that contains a translated
  transcription.

- Added `TransportParams.audio_in_passthrough`. If set (the default), incoming
  audio will be pushed downstream.

- Added `MCPClient`; a way to connect to MCP servers and use the MCP servers'
  tools.

- Added `Mem0 OSS`, along with Mem0 cloud support now the OSS version is also
  available.

### Changed

- `TransportParams.audio_mixer` now supports a string and also a dictionary to
  provide a mixer per destination. For example:

```python
TransportParams(
    audio_out_mixer={
        "track-1": SoundfileMixer(...),
        "track-2": SoundfileMixer(...),
        "track-N": SoundfileMixer(...),
    },
)
```

- The `STTMuteFilter` now mutes `InterimTranscriptionFrame` and
  `TranscriptionFrame` which allows the `STTMuteFilter` to be used in
  conjunction with transports that generate transcripts, e.g. `DailyTransport`.

- Function calls now receive a single parameter `FunctionCallParams` instead of
  `(function_name, tool_call_id, args, llm, context, result_callback)` which is
  now deprecated.

- Changed the user aggregator timeout for late transcriptions from 1.0s to 0.5s
  (`LLMUserAggregatorParams.aggregation_timeout`). Sometimes, the STT services
  might give us more than one transcription which could come after the user
  stopped speaking. We still want to include these additional transcriptions
  with the first one because it's part of the user turn. This is what this
  timeout is helpful with.

- Short utterances not detected by VAD while the bot is speaking are now
  ignored. This reduces the amount of bot interruptions significantly providing
  a more natural conversation experience.

- Updated `GladiaSTTService` to output a `TranslationFrame` when specifying a
  `translation` and `translation_config`.

- STT services now passthrough audio frames by default. This allows you to add
  audio recording without worrying about what's wrong in your pipeline when it
  doesn't work the first time.

- Input transports now always push audio downstream unless disabled with
  `TransportParams.audio_in_passthrough`. After many Pipecat releases, we
  realized this is the common use case. There are use cases where the input
  transport already provides STT and you also don't want recordings, in which
  case there's no need to push audio to the rest of the pipeline, but this is
  not a very common case.

- Added `RivaSegmentedSTTService`, which allows Riva offline/batch models, such
  as to be "canary-1b-asr" used in Pipecat.

### Deprecated

- Function calls with parameters
  `(function_name, tool_call_id, args, llm, context, result_callback)` are
  deprectated, use a single `FunctionCallParams` parameter instead.

- `TransportParams.camera_*` parameters are now deprecated, use
  `TransportParams.video_*` instead.

- `TransportParams.vad_enabled` parameter is now deprecated, use
  `TransportParams.audio_in_enabled` and `TransportParams.vad_analyzer` instead.

- `TransportParams.vad_audio_passthrough` parameter is now deprecated, use
  `TransportParams.audio_in_passthrough` instead.

- `ParakeetSTTService` is now deprecated, use `RivaSTTService` instead, which uses
  the model "parakeet-ctc-1.1b-asr" by default.

- `FastPitchTTSService` is now deprecated, use `RivaTTSService` instead, which uses
  the model "magpie-tts-multilingual" by default.

### Fixed

- Fixed an issue with `SimliVideoService` where the bot was continuously outputting
  audio, which prevents the `BotStoppedSpeakingFrame` from being emitted.

- Fixed an issue where `OpenAIRealtimeBetaLLMService` would add two assistant
  messages to the context.

- Fixed an issue with `GeminiMultimodalLiveLLMService` where the context
  contained tokens instead of words.

- Fixed an issue with HTTP Smart Turn handling, where the service returns a 500
  error. Previously, this would cause an unhandled exception. Now, a 500 error
  is treated as an incomplete response.

- Fixed a TTS services issue that could cause assistant output not to be
  aggregated to the context when also using `TTSSpeakFrame`s.

- Fixed an issue where the `SmartTurnMetricsData` was reporting 0ms for
  inference and processing time when using the `FalSmartTurnAnalyzer`.

### Other

- Added `examples/daily-custom-tracks` to show how to send and receive Daily
  custom tracks.

- Added `examples/daily-multi-translation` to showcase how to send multiple
  simulataneous translations with the same transport.

- Added 04 foundational examples for client/server transports. Also, renamed
  `29-livekit-audio-chat.py` to `04b-transports-livekit.py`.

- Added foundational example `13c-gladia-translation.py` showing how to use
  `TranscriptionFrame` and `TranslationFrame`.

## [0.0.65] - 2025-04-23 "Sant Jordi's release" 🌹📕

https://en.wikipedia.org/wiki/Saint_George%27s_Day_in_Catalonia

### Added

- Added automatic hangup logic to the Telnyx serializer. This feature hangs up
  the Telnyx call when an `EndFrame` or `CancelFrame` is received. It is
  enabled by default and is configurable via the `auto_hang_up` `InputParam`.

- Added a keepalive task to `GladiaSTTService` to prevent the websocket from
  disconnecting after 30 seconds of no audio input.

### Changed

- The `InputParams` for `ElevenLabsTTSService` and `ElevenLabsHttpTTSService`
  no longer require that `stability` and `similarity_boost` be set. You can
  individually set each param.

- In `TwilioFrameSerializer`, `call_sid` is Optional so as to avoid a breaking
  changed. `call_sid` is required to automatically hang up.

### Fixed

- Fixed an issue where `TwilioFrameSerializer` would send two hang up commands:
  one for the `EndFrame` and one for the `CancelFrame`.

## [0.0.64] - 2025-04-22

### Added

- Added automatic hangup logic to the Twilio serializer. This feature hangs up
  the Twilio call when an `EndFrame` or `CancelFrame` is received. It is
  enabled by default and is configurable via the `auto_hang_up` `InputParam`.

- Added `SmartTurnMetricsData`, which contains end-of-turn prediction metrics,
  to the `MetricsFrame`. Using `MetricsFrame`, you can now retrieve prediction
  confidence scores and processing time metrics from the smart turn analyzers.

- Added support for Application Default Credentials in Google services,
  `GoogleSTTService`, `GoogleTTSService`, and `GoogleVertexLLMService`.

- Added support for Smart Turn Detection via the `turn_analyzer` transport
  parameter. You can now choose between `HttpSmartTurnAnalyzer()` or
  `FalSmartTurnAnalyzer()` for remote inference or
  `LocalCoreMLSmartTurnAnalyzer()` for on-device inference using Core ML.

- `DeepgramTTSService` accepts `base_url` argument again, allowing you to
  connect to an on-prem service.

- Added `LLMUserAggregatorParams` and `LLMAssistantAggregatorParams` which allow
  you to control aggregator settings. You can now pass these arguments when
  creating aggregator pairs with `create_context_aggregator()`.

- Added `previous_text` context support to ElevenLabsHttpTTSService, improving
  speech consistency across sentences within an LLM response.

- Added word/timestamp pairs to `ElevenLabsHttpTTSService`.

- It is now possible to disable `SoundfileMixer` when created. You can then use
  `MixerEnableFrame` to dynamically enable it when necessary.

- Added `on_client_connected` and `on_client_disconnected` event handlers to
  the `DailyTransport` class. These handlers map to the same underlying Daily
  events as `on_participant_joined` and `on_participant_left`, respectively.
  This makes it easier to write a single bot pipeline that can also use other
  transports like `SmallWebRTCTransport` and `FastAPIWebsocketTransport`.

### Changed

- `GrokLLMService` now uses `grok-3-beta` as its default model.

- Daily's REST helpers now include an `eject_at_token_exp` param, which ejects
  the user when their token expires. This new parameter defaults to False.
  Also, the default value for `enable_prejoin_ui` changed to False and
  `eject_at_room_exp` changed to False.

- `OpenAILLMService` and `OpenPipeLLMService` now use `gpt-4.1` as their
  default model.

- `SoundfileMixer` constructor arguments need to be keywords.

### Deprecated

- `DeepgramSTTService` parameter `url` is now deprecated, use `base_url`
  instead.

### Removed

- Parameters `user_kwargs` and `assistant_kwargs` when creating a context
  aggregator pair using `create_context_aggregator()` have been removed. Use
  `user_params` and `assistant_params` instead.

### Fixed

- Fixed an issue that would cause TTS websocket-based services to not cleanup
  resources properly when disconnecting.

- Fixed a `TavusVideoService` issue that was causing audio choppiness.

- Fixed an issue in `SmallWebRTCTransport` where an error was thrown if the
  client did not create a video transceiver.

- Fixed an issue where LLM input parameters were not working and applied
  correctly in `GoogleVertexLLMService`, causing unexpected behavior during
  inference.

### Other

- Updated the `twilio-chatbot` example to use the auto-hangup feature.

## [0.0.63] - 2025-04-11

### Added

- Added media resolution control to `GeminiMultimodalLiveLLMService` with
  `GeminiMediaResolution` enum, allowing configuration of token usage for
  image processing (LOW: 64 tokens, MEDIUM: 256 tokens, HIGH: zoomed reframing
  with 256 tokens).

- Added Gemini's Voice Activity Detection (VAD) configuration to
  `GeminiMultimodalLiveLLMService` with `GeminiVADParams`, allowing fine
  control over speech detection sensitivity and timing, including:

  - Start sensitivity (how quickly speech is detected)
  - End sensitivity (how quickly turns end after pauses)
  - Prefix padding (milliseconds of audio to keep before speech is detected)
  - Silence duration (milliseconds of silence required to end a turn)

- Added comprehensive language support to `GeminiMultimodalLiveLLMService`,
  supporting over 30 languages via the `language` parameter, with proper
  mapping between Pipecat's `Language` enum and Gemini's language codes.

- Added support in `SmallWebRTCTransport` to detect when remote tracks are
  muted.

- Added support for image capture from a video stream to the
  `SmallWebRTCTransport`.

- Added a new iOS client option to the `SmallWebRTCTransport`
  **video-transform** example.

- Added new processors `ProducerProcessor` and `ConsumerProcessor`. The
  producer processor processes frames from the pipeline and decides whether the
  consumers should consume it or not. If so, the same frame that is received by
  the producer is sent to the consumer. There can be multiple consumers per
  producer. These processors can be useful to push frames from one part of a
  pipeline to a different one (e.g. when using `ParallelPipeline`).

- Improvements for the `SmallWebRTCTransport`:
  - Wait until the pipeline is ready before triggering the `connected` event.
  - Queue messages if the data channel is not ready.
  - Update the aiortc dependency to fix an issue where the 'video/rtx' MIME
    type was incorrectly handled as a codec retransmission.
  - Avoid initial video delays.

### Changed

- In `GeminiMultimodalLiveLLMService`, removed the `transcribe_model_audio`
  parameter in favor of Gemini Live's native output transcription support. Now
  text transcriptions are produced directly by the model. No configuration is
  required.

- Updated `GeminiMultimodalLiveLLMService`’s default `model` to
  `models/gemini-2.0-flash-live-001` and `base_url` to the `v1beta` websocket
  URL.

### Fixed

- Updated `daily-python` to 0.17.0 to fix an issue that was preventing to run on
  older platforms.

- Fixed an issue where `CartesiaTTSService`'s spell feature would result in
  the spelled word in the context appearing as "F,O,O,B,A,R" instead of
  "FOOBAR".

- Fixed an issue in the Azure TTS services where the language was being set
  incorrectly.

- Fixed `SmallWebRTCTransport` to support dynamic values for
  `TransportParams.audio_out_10ms_chunks`. Previously, it only worked with 20ms
  chunks.

- Fixed an issue with `GeminiMultimodalLiveLLMService` where the assistant
  context messages had no space between words.

- Fixed an issue where `LLMAssistantContextAggregator` would prevent a
  `BotStoppedSpeakingFrame` from moving through the pipeline.

## [0.0.62] - 2025-04-01 "An April Fools' release"

### Added

- Added `TransportParams.audio_out_10ms_chunks` parameter to allow controlling
  the amount of audio being sent by the output transport. It defaults to 4, so
  40ms audio chunks are sent.

- Added `QwenLLMService` for Qwen integration with an OpenAI-compatible
  interface. Added foundational example `14q-function-calling-qwen.py`.

- Added `Mem0MemoryService`. Mem0 is a self-improving memory layer for LLM
  applications. Learn more at: https://mem0.ai/.

- Added `WhisperSTTServiceMLX` for Whisper transcription on Apple Silicon.
  See example in `examples/foundational/13e-whisper-mlx.py`. Latency of
  completed transcription using Whisper large-v3-turbo on an M4 macbook is
  ~500ms.

- Added `SmallWebRTCTransport`, a new P2P WebRTC transport.

  - Created two examples in `p2p-webrtc`:
    - **video-transform**: Demonstrates sending and receiving audio/video with
      `SmallWebRTCTransport` using `TypeScript`. Includes video frame
      processing with OpenCV.
    - **voice-agent**: A minimal example of creating a voice agent with
      `SmallWebRTCTransport`.

- `GladiaSTTService` now have comprehensive support for the latest API config
  options, including model, language detection, preprocessing, custom
  vocabulary, custom spelling, translation, and message filtering options.

- Added `SmallWebRTCTransport`, a new P2P WebRTC transport.

  - Created two examples in `p2p-webrtc`:
    - **video-transform**: Demonstrates sending and receiving audio/video with
      `SmallWebRTCTransport` using `TypeScript`. Includes video frame
      processing with OpenCV.
    - **voice-agent**: A minimal example of creating a voice agent with
      `SmallWebRTCTransport`.

- Added support to `ProtobufFrameSerializer` to send the messages from
  `TransportMessageFrame` and `TransportMessageUrgentFrame`.

- Added support for a new TTS service, `PiperTTSService`.
  (see https://github.com/rhasspy/piper/)

- It is now possible to tell whether `UserStartedSpeakingFrame` or
  `UserStoppedSpeakingFrame` have been generated because of emulation frames.

### Changed

- `FunctionCallResultFrame`a are now system frames. This is to prevent function
  call results to be discarded during interruptions.

- Pipecat services have been reorganized into packages. Each package can have
  one or more of the following modules (in the future new module names might be
  needed) depending on the services implemented:

  - image: for image generation services
  - llm: for LLM services
  - memory: for memory services
  - stt: for Speech-To-Text services
  - tts: for Text-To-Speech services
  - video: for video generation services
  - vision: for video recognition services

- Base classes for AI services have been reorganized into modules. They can now
  be found in
  `pipecat.services.[ai_service,image_service,llm_service,stt_service,vision_service]`.

- `GladiaSTTService` now uses the `solaria-1` model by default. Other params
  use Gladia's default values. Added support for more language codes.

### Deprecated

- All Pipecat services imports have been deprecated and a warning will be shown
  when using the old import. The new import should be
  `pipecat.services.[service].[image,llm,memory,stt,tts,video,vision]`. For
  example, `from pipecat.services.openai.llm import OpenAILLMService`.

- Import for AI services base classes from `pipecat.services.ai_services` is now
  deprecated, use one of
  `pipecat.services.[ai_service,image_service,llm_service,stt_service,vision_service]`.

- Deprecated the `language` parameter in `GladiaSTTService.InputParams` in
  favor of `language_config`, which better aligns with Gladia's API.

- Deprecated using `GladiaSTTService.InputParams` directly. Use the new
  `GladiaInputParams` class instead.

### Fixed

- Fixed a `FastAPIWebsocketTransport` and `WebsocketClientTransport` issue that
  would cause the transport to be closed prematurely, preventing the internally
  queued audio to be sent. The same issue could also cause an infinite loop
  while using an output mixer and when sending an `EndFrame`, preventing the bot
  to finish.

- Fixed an issue that could cause the `TranscriptionUpdateFrame` being pushed
  because of an interruption to be discarded.

- Fixed an issue that would cause `SegmentedSTTService` based services
  (e.g. `OpenAISTTService`) to try to transcribe non-spoken audio, causing
  invalid transcriptions.

- Fixed an issue where `GoogleTTSService` was emitting two `TTSStoppedFrames`.

### Performance

- Output transports now send 40ms audio chunks instead of 20ms. This should
  improve performance.

- `BotSpeakingFrame`s are now sent every 200ms. If the output transport audio chunks
  are higher than 200ms then they will be sent at every audio chunk.

### Other

- Added foundational example `37-mem0.py` demonstrating how to use the
  `Mem0MemoryService`.

- Added foundational example `13e-whisper-mlx.py` demonstrating how to use the
  `WhisperSTTServiceMLX`.

## [0.0.61] - 2025-03-26

### Added

- Added a new frame, `LLMSetToolChoiceFrame`, which provides a mechanism
  for modifying the `tool_choice` in the context.

- Added `GroqTTSService` which provides text-to-speech functionality using
  Groq's API.

- Added support in `DailyTransport` for updating remote participants'
  `canReceive` permission via the `update_remote_participants()` method, by
  bumping the daily-python dependency to >= 0.16.0.

- ElevenLabs TTS services now support a sample rate of 8000.

- Added support for `instructions` in `OpenAITTSService`.

- Added support for `base_url` in `OpenAIImageGenService` and
  `OpenAITTSService`.

### Fixed

- Fixed an issue in `RTVIObserver` that prevented handling of Google LLM
  context messages. The observer now processes both OpenAI-style and
  Google-style contexts.

- Fixed an issue in Daily involving switching virtual devices, by bumping the
  daily-python dependency to >= 0.16.1.

- Fixed a `GoogleAssistantContextAggregator` issue where function calls
  placeholders where not being updated when then function call result was
  different from a string.

- Fixed an issue that would cause `LLMAssistantContextAggregator` to block
  processing more frames while processing a function call result.

- Fixed an issue where the `RTVIObserver` would report two bot started and
  stopped speaking events for each bot turn.

- Fixed an issue in `UltravoxSTTService` that caused improper audio processing
  and incorrect LLM frame output.

### Other

- Added `examples/foundational/07x-interruptible-local.py` to show how a local
  transport can be used.

## [0.0.60] - 2025-03-20

### Added

- Added `default_headers` parameter to `BaseOpenAILLMService` constructor.

### Changed

- Rollback to `deepgram-sdk` 3.8.0 since 3.10.1 was causing connections issues.

- Changed the default `InputAudioTranscription` model to `gpt-4o-transcribe`
  for `OpenAIRealtimeBetaLLMService`.

### Other

- Update the `19-openai-realtime-beta.py` and `19a-azure-realtime-beta.py`
  examples to use the FunctionSchema format.

## [0.0.59] - 2025-03-20

### Added

- When registering a function call it is now possible to indicate if you want
  the function call to be cancelled if there's a user interruption via
  `cancel_on_interruption` (defaults to False). This is now possible because
  function calls are executed concurrently.

- Added support for detecting idle pipelines. By default, if no activity has
  been detected during 5 minutes, the `PipelineTask` will be automatically
  cancelled. It is possible to override this behavior by passing
  `cancel_on_idle_timeout=False`. It is also possible to change the default
  timeout with `idle_timeout_secs` or the frames that prevent the pipeline from
  being idle with `idle_timeout_frames`. Finally, an `on_idle_timeout` event
  handler will be triggered if the idle timeout is reached (whether the pipeline
  task is cancelled or not).

- Added `FalSTTService`, which provides STT for Fal's Wizper API.

- Added a `reconnect_on_error` parameter to websocket-based TTS services as well
  as a `on_connection_error` event handler. The `reconnect_on_error` indicates
  whether the TTS service should reconnect on error. The `on_connection_error`
  will always get called if there's any error no matter the value of
  `reconnect_on_error`. This allows, for example, to fallback to a different TTS
  provider if something goes wrong with the current one.

- Added new `SkipTagsAggregator` that extends `BaseTextAggregator` to aggregate
  text and skips end of sentence matching if aggregated text is between
  start/end tags.

- Added new `PatternPairAggregator` that extends `BaseTextAggregator` to
  identify content between matching pattern pairs in streamed text. This allows
  for detection and processing of structured content like XML-style tags that
  may span across multiple text chunks or sentence boundaries.

- Added new `BaseTextAggregator`. Text aggregators are used by the TTS service
  to aggregate LLM tokens and decide when the aggregated text should be pushed
  to the TTS service. They also allow for the text to be manipulated while it's
  being aggregated. A text aggregator can be passed via `text_aggregator` to the
  TTS service.

- Added new `sample_rate` constructor parameter to `TavusVideoService` to allow
  changing the output sample rate.

- Added new `NeuphonicTTSService`.
  (see https://neuphonic.com)

- Added new `UltravoxSTTService`.
  (see https://github.com/fixie-ai/ultravox)

- Added `on_frame_reached_upstream` and `on_frame_reached_downstream` event
  handlers to `PipelineTask`. Those events will be called when a frame reaches
  the beginning or end of the pipeline respectively. Note that by default, the
  event handlers will not be called unless a filter is set with
  `PipelineTask.set_reached_upstream_filter()` or
  `PipelineTask.set_reached_downstream_filter()`.

- Added support for Chirp voices in `GoogleTTSService`.

- Added a `flush_audio()` method to `FishTTSService` and `LmntTTSService`.

- Added a `set_language` convenience method for `GoogleSTTService`, allowing
  you to set a single language. This is in addition to the `set_languages`
  method which allows you to set a list of languages.

- Added `on_user_turn_audio_data` and `on_bot_turn_audio_data` to
  `AudioBufferProcessor`. This gives the ability to grab the audio of only that
  turn for both the user and the bot.

- Added new base class `BaseObject` which is now the base class of
  `FrameProcessor`, `PipelineRunner`, `PipelineTask` and `BaseTransport`. The
  new `BaseObject` adds supports for event handlers.

- Added support for a unified format for specifying function calling across all
  LLM services.

```python
  weather_function = FunctionSchema(
      name="get_current_weather",
      description="Get the current weather",
      properties={
          "location": {
              "type": "string",
              "description": "The city and state, e.g. San Francisco, CA",
          },
          "format": {
              "type": "string",
              "enum": ["celsius", "fahrenheit"],
              "description": "The temperature unit to use. Infer this from the user's location.",
          },
      },
      required=["location"],
  )
  tools = ToolsSchema(standard_tools=[weather_function])
```

- Added `speech_threshold` parameter to `GladiaSTTService`.

- Allow passing user (`user_kwargs`) and assistant (`assistant_kwargs`) context
  aggregator parameters when using `create_context_aggregator()`. The values are
  passed as a mapping that will then be converted to arguments.

- Added `speed` as an `InputParam` for both `ElevenLabsTTSService` and
  `ElevenLabsHttpTTSService`.

- Added new `LLMFullResponseAggregator` to aggregate full LLM completions. At
  every completion the `on_completion` event handler is triggered.

- Added a new frame, `RTVIServerMessageFrame`, and RTVI message
  `RTVIServerMessage` which provides a generic mechanism for sending custom
  messages from server to client. The `RTVIServerMessageFrame` is processed by
  the `RTVIObserver` and will be delivered to the client's `onServerMessage`
  callback or `ServerMessage` event.

- Added `GoogleLLMOpenAIBetaService` for Google LLM integration with an
  OpenAI-compatible interface. Added foundational example
  `14o-function-calling-gemini-openai-format.py`.

- Added `AzureRealtimeBetaLLMService` to support Azure's OpeanAI Realtime API. Added
  foundational example `19a-azure-realtime-beta.py`.

- Introduced `GoogleVertexLLMService`, a new class for integrating with Vertex AI
  Gemini models. Added foundational example
  `14p-function-calling-gemini-vertex-ai.py`.

- Added support in `OpenAIRealtimeBetaLLMService` for a slate of new features:

  - The `'gpt-4o-transcribe'` input audio transcription model, along
    with new `language` and `prompt` options specific to that model.
  - The `input_audio_noise_reduction` session property.

    ```python
    session_properties = SessionProperties(
        # ...
        input_audio_noise_reduction=InputAudioNoiseReduction(
            type="near_field"  # also supported: "far_field"
        )
        # ...
    )
    ```

  - The `'semantic_vad'` `turn_detection` session property value, a more
    sophisticated model for detecting when the user has stopped speaking.
  - `on_conversation_item_created` and `on_conversation_item_updated`
    events to `OpenAIRealtimeBetaLLMService`.

    ```python
    @llm.event_handler("on_conversation_item_created")
    async def on_conversation_item_created(llm, item_id, item):
      # ...

    @llm.event_handler("on_conversation_item_updated")
    async def on_conversation_item_updated(llm, item_id, item):
      # `item` may not always be available here
      # ...
    ```

  - The `retrieve_conversation_item(item_id)` method for introspecting a
    conversation item on the server.

    ```python
    item = await llm.retrieve_conversation_item(item_id)
    ```

### Changed

- Updated `OpenAISTTService` to use `gpt-4o-transcribe` as the default
  transcription model.

- Updated `OpenAITTSService` to use `gpt-4o-mini-tts` as the default TTS model.

- Function calls are now executed in tasks. This means that the pipeline will
  not be blocked while the function call is being executed.

- ⚠️ `PipelineTask` will now be automatically cancelled if no bot activity is
  happening in the pipeline. There are a few settings to configure this
  behavior, see `PipelineTask` documentation for more details.

- All event handlers are now executed in separate tasks in order to prevent
  blocking the pipeline. It is possible that event handlers take some time to
  execute in which case the pipeline would be blocked waiting for the event
  handler to complete.

- Updated `TranscriptProcessor` to support text output from
  `OpenAIRealtimeBetaLLMService`.

- `OpenAIRealtimeBetaLLMService` and `GeminiMultimodalLiveLLMService` now push
  a `TTSTextFrame`.

- Updated the default mode for `CartesiaTTSService` and
  `CartesiaHttpTTSService` to `sonic-2`.

### Deprecated

- Passing a `start_callback` to `LLMService.register_function()` is now
  deprecated, simply move the code from the start callback to the function call.

- `TTSService` parameter `text_filter` is now deprecated, use `text_filters`
  instead which is now a list. This allows passing multiple filters that will be
  executed in order.

### Removed

- Removed deprecated `audio.resample_audio()`, use `create_default_resampler()`
  instead.

- Removed deprecated`stt_service` parameter from `STTMuteFilter`.

- Removed deprecated RTVI processors, use an `RTVIObserver` instead.

- Removed deprecated `AWSTTSService`, use `PollyTTSService` instead.

- Removed deprecated field `tier` from `DailyTranscriptionSettings`, use `model`
  instead.

- Removed deprecated `pipecat.vad` package, use `pipecat.audio.vad` instead.

### Fixed

- Fixed an assistant aggregator issue that could cause assistant text to be
  split into multiple chunks during function calls.

- Fixed an assistant aggregator issue that was causing assistant text to not be
  added to the context during function calls. This could lead to duplications.

- Fixed a `SegmentedSTTService` issue that was causing audio to be sent
  prematurely to the STT service. Instead of analyzing the volume in this
  service we rely on VAD events which use both VAD and volume.

- Fixed a `GeminiMultimodalLiveLLMService` issue that was causing messages to be
  duplicated in the context when pushing `LLMMessagesAppendFrame` frames.

- Fixed an issue with `SegmentedSTTService` based services
  (e.g. `GroqSTTService`) that was not allow audio to pass-through downstream.

- Fixed a `CartesiaTTSService` and `RimeTTSService` issue that would consider
  text between spelling out tags end of sentence.

- Fixed a `match_endofsentence` issue that would result in floating point
  numbers to be considered an end of sentence.

- Fixed a `match_endofsentence` issue that would result in emails to be
  considered an end of sentence.

- Fixed an issue where the RTVI message `disconnect-bot` was pushing an
  `EndFrame`, resulting in the pipeline not shutting down. It now pushes an
  `EndTaskFrame` upstream to shutdown the pipeline.

- Fixed an issue with the `GoogleSTTService` where stream timeouts during
  periods of inactivity were causing connection failures. The service now
  properly detects timeout errors and handles reconnection gracefully,
  ensuring continuous operation even after periods of silence or when using an
  `STTMuteFilter`.

- Fixed an issue in `RimeTTSService` where the last line of text sent didn't
  result in an audio output being generated.

- Fixed `OpenAIRealtimeBetaLLMService` by adding proper handling for:
  - The `conversation.item.input_audio_transcription.delta` server message,
    which was added server-side at some point and not handled client-side.
  - Errors reported by the `response.done` server message.

### Other

- Add foundational example `07w-interruptible-fal.py`, showing `FalSTTService`.

- Added a new Ultravox example
  `examples/foundational/07u-interruptible-ultravox.py`.

- Added new Neuphonic examples
  `examples/foundational/07v-interruptible-neuphonic.py` and
  `examples/foundational/07v-interruptible-neuphonic-http.py`.

- Added a new example `examples/foundational/36-user-email-gathering.py` to show
  how to gather user emails. The example uses's Cartesia's `<spell></spell>`
  tags and Rime `spell()` function to spell out the emails for confirmation.

- Update the `34-audio-recording.py` example to include an STT processor.

- Added foundational example `35-voice-switching.py` showing how to use the new
  `PatternPairAggregator`. This example shows how to encode information for the
  LLM to instruct TTS voice changes, but this can be used to encode any
  information into the LLM response, which you want to parse and use in other
  parts of your application.

- Added a Pipecat Cloud deployment example to the `examples` directory.

- Removed foundational examples 28b and 28c as the TranscriptProcessor no
  longer has an LLM depedency. Renamed foundational example 28a to
  `28-transcript-processor.py`.

## [0.0.58] - 2025-02-26

### Added

- Added track-specific audio event `on_track_audio_data` to
  `AudioBufferProcessor` for accessing separate input and output audio tracks.

- Pipecat version will now be logged on every application startup. This will
  help us identify what version we are running in case of any issues.

- Added a new `StopFrame` which can be used to stop a pipeline task while
  keeping the frame processors running. The frame processors could then be used
  in a different pipeline. The difference between a `StopFrame` and a
  `StopTaskFrame` is that, as with `EndFrame` and `EndTaskFrame`, the
  `StopFrame` is pushed from the task and the `StopTaskFrame` is pushed upstream
  inside the pipeline by any processor.

- Added a new `PipelineTask` parameter `observers` that replaces the previous
  `PipelineParams.observers`.

- Added a new `PipelineTask` parameter `check_dangling_tasks` to enable or
  disable checking for frame processors' dangling tasks when the Pipeline
  finishes running.

- Added new `on_completion_timeout` event for LLM services (all OpenAI-based
  services, Anthropic and Google). Note that this event will only get triggered
  if LLM timeouts are setup and if the timeout was reached. It can be useful to
  retrigger another completion and see if the timeout was just a blip.

- Added new log observers `LLMLogObserver` and `TranscriptionLogObserver` that
  can be useful for debugging your pipelines.

- Added `room_url` property to `DailyTransport`.

- Added `addons` argument to `DeepgramSTTService`.

- Added `exponential_backoff_time()` to `utils.network` module.

### Changed

- ⚠️ `PipelineTask` now requires keyword arguments (except for the first one for
  the pipeline).

- Updated `PlayHTHttpTTSService` to take a `voice_engine` and `protocol` input
  in the constructor. The previous method of providing a `voice_engine` input
  that contains the engine and protocol is deprecated by PlayHT.

- The base `TTSService` class now strips leading newlines before sending text
  to the TTS provider. This change is to solve issues where some TTS providers,
  like Azure, would not output text due to newlines.

- `GrokLLMSService` now uses `grok-2` as the default model.

- `AnthropicLLMService` now uses `claude-3-7-sonnet-20250219` as the default
  model.

- `RimeHttpTTSService` needs an `aiohttp.ClientSession` to be passed to the
  constructor as all the other HTTP-based services.

- `RimeHttpTTSService` doesn't use a default voice anymore.

- `DeepgramSTTService` now uses the new `nova-3` model by default. If you want
  to use the previous model you can pass `LiveOptions(model="nova-2-general")`.
  (see https://deepgram.com/learn/introducing-nova-3-speech-to-text-api)

```python
stt = DeepgramSTTService(..., live_options=LiveOptions(model="nova-2-general"))
```

### Deprecated

- `PipelineParams.observers` is now deprecated, you the new `PipelineTask`
  parameter `observers`.

### Removed

- Remove `TransportParams.audio_out_is_live` since it was not being used at all.

### Fixed

- Fixed an issue that would cause undesired interruptions via
  `EmulateUserStartedSpeakingFrame`.

- Fixed a `GoogleLLMService` that was causing an exception when sending inline
  audio in some cases.

- Fixed an `AudioContextWordTTSService` issue that would cause an `EndFrame` to
  disconnect from the TTS service before audio from all the contexts was
  received. This affected services like Cartesia and Rime.

- Fixed an issue that was not allowing to pass an `OpenAILLMContext` to create
  `GoogleLLMService`'s context aggregators.

- Fixed a `ElevenLabsTTSService`, `FishAudioTTSService`, `LMNTTTSService` and
  `PlayHTTTSService` issue that was resulting in audio requested before an
  interruption being played after an interruption.

- Fixed `match_endofsentence` support for ellipses.

- Fixed an issue where `EndTaskFrame` was not triggering
  `on_client_disconnected` or closing the WebSocket in FastAPI.

- Fixed an issue in `DeepgramSTTService` where the `sample_rate` passed to the
  `LiveOptions` was not being used, causing the service to use the default
  sample rate of pipeline.

- Fixed a context aggregator issue that would not append the LLM text response
  to the context if a function call happened in the same LLM turn.

- Fixed an issue that was causing HTTP TTS services to push `TTSStoppedFrame`
  more than once.

- Fixed a `FishAudioTTSService` issue where `TTSStoppedFrame` was not being
  pushed.

- Fixed an issue that `start_callback` was not invoked for some LLM services.

- Fixed an issue that would cause `DeepgramSTTService` to stop working after an
  error occurred (e.g. sudden network loss). If the network recovered we would
  not reconnect.

- Fixed a `STTMuteFilter` issue that would not mute user audio frames causing
  transcriptions to be generated by the STT service.

### Other

- Added Gemini support to `examples/phone-chatbot`.

- Added foundational example `34-audio-recording.py` showing how to use the
  AudioBufferProcessor callbacks to save merged and track recordings.

## [0.0.57] - 2025-02-14

### Added

- Added new `AudioContextWordTTSService`. This is a TTS base class for TTS
  services that handling multiple separate audio requests.

- Added new frames `EmulateUserStartedSpeakingFrame` and
  `EmulateUserStoppedSpeakingFrame` which can be used to emulated VAD behavior
  without VAD being present or not being triggered.

- Added a new `audio_in_stream_on_start` field to `TransportParams`.

- Added a new method `start_audio_in_streaming` in the `BaseInputTransport`.

  - This method should be used to start receiving the input audio in case the
    field `audio_in_stream_on_start` is set to `false`.

- Added support for the `RTVIProcessor` to handle buffered audio in `base64`
  format, converting it into InputAudioRawFrame for transport.

- Added support for the `RTVIProcessor` to trigger `start_audio_in_streaming`
  only after the `client-ready` message.

- Added new `MUTE_UNTIL_FIRST_BOT_COMPLETE` strategy to `STTMuteStrategy`. This
  strategy starts muted and remains muted until the first bot speech completes,
  ensuring the bot's first response cannot be interrupted. This complements the
  existing `FIRST_SPEECH` strategy which only mutes during the first detected
  bot speech.

- Added support for Google Cloud Speech-to-Text V2 through `GoogleSTTService`.

- Added `RimeTTSService`, a new `WordTTSService`. Updated the foundational
  example `07q-interruptible-rime.py` to use `RimeTTSService`.

- Added support for Groq's Whisper API through the new `GroqSTTService` and
  OpenAI's Whisper API through the new `OpenAISTTService`. Introduced a new
  base class `BaseWhisperSTTService` to handle common Whisper API
  functionality.

- Added `PerplexityLLMService` for Perplexity NIM API integration, with an
  OpenAI-compatible interface. Also, added foundational example
  `14n-function-calling-perplexity.py`.

- Added `DailyTransport.update_remote_participants()`. This allows you to update
  remote participant's settings, like their permissions or which of their
  devices are enabled. Requires that the local participant have participant
  admin permission.

### Changed

- We don't consider a colon `:` and end of sentence any more.

- Updated `DailyTransport` to respect the `audio_in_stream_on_start` field,
  ensuring it only starts receiving the audio input if it is enabled.

- Updated `FastAPIWebsocketOutputTransport` to send `TransportMessageFrame` and
  `TransportMessageUrgentFrame` to the serializer.

- Updated `WebsocketServerOutputTransport` to send `TransportMessageFrame` and
  `TransportMessageUrgentFrame` to the serializer.

- Enhanced `STTMuteConfig` to validate strategy combinations, preventing
  `MUTE_UNTIL_FIRST_BOT_COMPLETE` and `FIRST_SPEECH` from being used together
  as they handle first bot speech differently.

- Updated foundational example `07n-interruptible-google.py` to use all Google
  services.

- `RimeHttpTTSService` now uses the `mistv2` model by default.

- Improved error handling in `AzureTTSService` to properly detect and log
  synthesis cancellation errors.

- Enhanced `WhisperSTTService` with full language support and improved model
  documentation.

- Updated foundation example `14f-function-calling-groq.py` to use
  `GroqSTTService` for transcription.

- Updated `GroqLLMService` to use `llama-3.3-70b-versatile` as the default
  model.

- `RTVIObserver` doesn't handle `LLMSearchResponseFrame` frames anymore. For
  now, to handle those frames you need to create a `GoogleRTVIObserver`
  instead.

### Deprecated

- `STTMuteFilter` constructor's `stt_service` parameter is now deprecated and
  will be removed in a future version. The filter now manages mute state
  internally instead of querying the STT service.

- `RTVI.observer()` is now deprecated, instantiate an `RTVIObserver` directly
  instead.

- All RTVI frame processors (e.g. `RTVISpeakingProcessor`,
  `RTVIBotLLMProcessor`) are now deprecated, instantiate an `RTVIObserver`
  instead.

### Fixed

- Fixed a `FalImageGenService` issue that was causing the event loop to be
  blocked while loading the downloadded image.

- Fixed a `CartesiaTTSService` service issue that would cause audio overlapping
  in some cases.

- Fixed a websocket-based service issue (e.g. `CartesiaTTSService`) that was
  preventing a reconnection after the server disconnected cleanly, which was
  causing an inifite loop instead.

- Fixed a `BaseOutputTransport` issue that was causing upstream frames to no be
  pushed upstream.

- Fixed multiple issue where user transcriptions where not being handled
  properly. It was possible for short utterances to not trigger VAD which would
  cause user transcriptions to be ignored. It was also possible for one or more
  transcriptions to be generated after VAD in which case they would also be
  ignored.

- Fixed an issue that was causing `BotStoppedSpeakingFrame` to be generated too
  late. This could then cause issues unblocking `STTMuteFilter` later than
  desired.

- Fixed an issue that was causing `AudioBufferProcessor` to not record
  synchronized audio.

- Fixed an `RTVI` issue that was causing `bot-tts-text` messages to be sent
  before being processed by the output transport.

- Fixed an issue[#1192] in 11labs where we are trying to reconnect/disconnect
  the websocket connection even when the connection is already closed.

- Fixed an issue where `has_regular_messages` condition was always true in
  `GoogleLLMContext` due to `Part` having `function_call` & `function_response`
  with `None` values.

### Other

- Added new `instant-voice` example. This example showcases how to enable
  instant voice communication as soon as a user connects.

- Added new `local-input-select-stt` example. This examples allows you to play
  with local audio inputs by slecting them through a nice text interface.

## [0.0.56] - 2025-02-06

### Changed

- Use `gemini-2.0-flash-001` as the default model for `GoogleLLMSerivce`.

- Improved foundational examples 22b, 22c, and 22d to support function calling.
  With these base examples, `FunctionCallInProgressFrame` and
  `FunctionCallResultFrame` will no longer be blocked by the gates.

### Fixed

- Fixed a `TkLocalTransport` and `LocalAudioTransport` issues that was causing
  errors on cleanup.

- Fixed an issue that was causing `tests.utils` import to fail because of
  logging setup.

- Fixed a `SentryMetrics` issue that was preventing any metrics to be sent to
  Sentry and also was preventing from metrics frames to be pushed to the
  pipeline.

- Fixed an issue in `BaseOutputTransport` where incoming audio would not be
  resampled to the desired output sample rate.

- Fixed an issue with the `TwilioFrameSerializer` and `TelnyxFrameSerializer`
  where `twilio_sample_rate` and `telnyx_sample_rate` were incorrectly
  initialized to `audio_in_sample_rate`. Those values currently default to 8000
  and should be set manually from the serializer constructor if a different
  value is needed.

### Other

- Added a new `sentry-metrics` example.

## [0.0.55] - 2025-02-05

### Added

- Added a new `start_metadata` field to `PipelineParams`. The provided metadata
  will be set to the initial `StartFrame` being pushed from the `PipelineTask`.

- Added new fields to `PipelineParams` to control audio input and output sample
  rates for the whole pipeline. This allows controlling sample rates from a
  single place instead of having to specify sample rates in each
  service. Setting a sample rate to a service is still possible and will
  override the value from `PipelineParams`.

- Introduce audio resamplers (`BaseAudioResampler`). This is just a base class
  to implement audio resamplers. Currently, two implementations are provided
  `SOXRAudioResampler` and `ResampyResampler`. A new
  `create_default_resampler()` has been added (replacing the now deprecated
  `resample_audio()`).

- It is now possible to specify the asyncio event loop that a `PipelineTask` and
  all the processors should run on by passing it as a new argument to the
  `PipelineRunner`. This could allow running pipelines in multiple threads each
  one with its own event loop.

- Added a new `utils.TaskManager`. Instead of a global task manager we now have
  a task manager per `PipelineTask`. In the previous version the task manager
  was global, so running multiple simultaneous `PipelineTask`s could result in
  dangling task warnings which were not actually true. In order, for all the
  processors to know about the task manager, we pass it through the
  `StartFrame`. This means that processors should create tasks when they receive
  a `StartFrame` but not before (because they don't have a task manager yet).

- Added `TelnyxFrameSerializer` to support Telnyx calls. A full running example
  has also been added to `examples/telnyx-chatbot`.

- Allow pushing silence audio frames before `TTSStoppedFrame`. This might be
  useful for testing purposes, for example, passing bot audio to an STT service
  which usually needs additional audio data to detect the utterance stopped.

- `TwilioSerializer` now supports transport message frames. With this we can
  create Twilio emulators.

- Added a new transport: `WebsocketClientTransport`.

- Added a `metadata` field to `Frame` which makes it possible to pass custom
  data to all frames.

- Added `test/utils.py` inside of pipecat package.

### Changed

- `GatedOpenAILLMContextAggregator` now require keyword arguments. Also, a new
  `start_open` argument has been added to set the initial state of the gate.

- Added `organization` and `project` level authentication to
  `OpenAILLMService`.

- Improved the language checking logic in `ElevenLabsTTSService` and
  `ElevenLabsHttpTTSService` to properly handle language codes based on model
  compatibility, with appropriate warnings when language codes cannot be
  applied.

- Updated `GoogleLLMContext` to support pushing `LLMMessagesUpdateFrame`s that
  contain a combination of function calls, function call responses, system
  messages, or just messages.

- `InputDTMFFrame` is now based on `DTMFFrame`. There's also a new
  `OutputDTMFFrame` frame.

### Deprecated

- `resample_audio()` is now deprecated, use `create_default_resampler()`
  instead.

### Removed

- `AudioBufferProcessor.reset_audio_buffers()` has been removed, use
  `AudioBufferProcessor.start_recording()` and
  `AudioBufferProcessor.stop_recording()` instead.

### Fixed

- Fixed a `AudioBufferProcessor` that would cause crackling in some recordings.

- Fixed an issue in `AudioBufferProcessor` where user callback would not be
  called on task cancellation.

- Fixed an issue in `AudioBufferProcessor` that would cause wrong silence
  padding in some cases.

- Fixed an issue where `ElevenLabsTTSService` messages would return a 1009
  websocket error by increasing the max message size limit to 16MB.

- Fixed a `DailyTransport` issue that would cause events to be triggered before
  join finished.

- Fixed a `PipelineTask` issue that was preventing processors to be cleaned up
  after cancelling the task.

- Fixed an issue where queuing a `CancelFrame` to a pipeline task would not
  cause the task to finish. However, using `PipelineTask.cancel()` is still the
  recommended way to cancel a task.

### Other

- Improved Unit Test `run_test()` to use `PipelineTask` and
  `PipelineRunner`. There's now also some control around `StartFrame` and
  `EndFrame`. The `EndTaskFrame` has been removed since it doesn't seem
  necessary with this new approach.

- Updated `twilio-chatbot` with a few new features: use 8000 sample rate and
  avoid resampling, a new client useful for stress testing and testing locally
  without the need to make phone calls. Also, added audio recording on both the
  client and the server to make sure the audio sounds good.

- Updated examples to use `task.cancel()` to immediately exit the example when a
  participant leaves or disconnects, instead of pushing an `EndFrame`. Pushing
  an `EndFrame` causes the bot to run through everything that is internally
  queued (which could take some seconds). Note that using `task.cancel()` might
  not always be the best option and pushing an `EndFrame` could still be
  desirable to make sure all the pipeline is flushed.

## [0.0.54] - 2025-01-27

### Added

- In order to create tasks in Pipecat frame processors it is now recommended to
  use `FrameProcessor.create_task()` (which uses the new
  `utils.asyncio.create_task()`). It takes care of uncaught exceptions, task
  cancellation handling and task management. To cancel or wait for a task there
  is `FrameProcessor.cancel_task()` and `FrameProcessor.wait_for_task()`. All of
  Pipecat processors have been updated accordingly. Also, when a pipeline runner
  finishes, a warning about dangling tasks might appear, which indicates if any
  of the created tasks was never cancelled or awaited for (using these new
  functions).

- It is now possible to specify the period of the `PipelineTask` heartbeat
  frames with `heartbeats_period_secs`.

- Added `DailyMeetingTokenProperties` and `DailyMeetingTokenParams` Pydantic models
  for meeting token creation in `get_token` method of `DailyRESTHelper`.

- Added `enable_recording` and `geo` parameters to `DailyRoomProperties`.

- Added `RecordingsBucketConfig` to `DailyRoomProperties` to upload recordings
  to a custom AWS bucket.

### Changed

- Enhanced `UserIdleProcessor` with retry functionality and control over idle
  monitoring via new callback signature `(processor, retry_count) -> bool`.
  Updated the `17-detect-user-idle.py` to show how to use the `retry_count`.

- Add defensive error handling for `OpenAIRealtimeBetaLLMService`'s audio
  truncation. Audio truncation errors during interruptions now log a warning
  and allow the session to continue instead of throwing an exception.

- Modified `TranscriptProcessor` to use TTS text frames for more accurate assistant
  transcripts. Assistant messages are now aggregated based on bot speaking boundaries
  rather than LLM context, providing better handling of interruptions and partial
  utterances.

- Updated foundational examples `28a-transcription-processor-openai.py`,
  `28b-transcript-processor-anthropic.py`, and
  `28c-transcription-processor-gemini.py` to use the updated
  `TranscriptProcessor`.

### Fixed

- Fixed an `GeminiMultimodalLiveLLMService` issue that was preventing the user
  to push initial LLM assistant messages (using `LLMMessagesAppendFrame`).

- Added missing `FrameProcessor.cleanup()` calls to `Pipeline`,
  `ParallelPipeline` and `UserIdleProcessor`.

- Fixed a type error when using `voice_settings` in `ElevenLabsHttpTTSService`.

- Fixed an issue where `OpenAIRealtimeBetaLLMService` function calling resulted
  in an error.

- Fixed an issue in `AudioBufferProcessor` where the last audio buffer was not
  being processed, in cases where the `_user_audio_buffer` was smaller than the
  buffer size.

### Performance

- Replaced audio resampling library `resampy` with `soxr`. Resampling a 2:21s
  audio file from 24KHz to 16KHz took 1.41s with `resampy` and 0.031s with
  `soxr` with similar audio quality.

### Other

- Added initial unit test infrastructure.

## [0.0.53] - 2025-01-18

### Added

- Added `ElevenLabsHttpTTSService` which uses EleveLabs' HTTP API instead of the
  websocket one.

- Introduced pipeline frame observers. Observers can view all the frames that go
  through the pipeline without the need to inject processors in the
  pipeline. This can be useful, for example, to implement frame loggers or
  debuggers among other things. The example
  `examples/foundational/30-observer.py` shows how to add an observer to a
  pipeline for debugging.

- Introduced heartbeat frames. The pipeline task can now push periodic
  heartbeats down the pipeline when `enable_heartbeats=True`. Heartbeats are
  system frames that are supposed to make it all the way to the end of the
  pipeline. When a heartbeat frame is received the traversing time (i.e. the
  time it took to go through the whole pipeline) will be displayed (with TRACE
  logging) otherwise a warning will be shown. The example
  `examples/foundational/31-heartbeats.py` shows how to enable heartbeats and
  forces warnings to be displayed.

- Added `LLMTextFrame` and `TTSTextFrame` which should be pushed by LLM and TTS
  services respectively instead of `TextFrame`s.

- Added `OpenRouter` for OpenRouter integration with an OpenAI-compatible
  interface. Added foundational example `14m-function-calling-openrouter.py`.

- Added a new `WebsocketService` based class for TTS services, containing
  base functions and retry logic.

- Added `DeepSeekLLMService` for DeepSeek integration with an OpenAI-compatible
  interface. Added foundational example `14l-function-calling-deepseek.py`.

- Added `FunctionCallResultProperties` dataclass to provide a structured way to
  control function call behavior, including:

  - `run_llm`: Controls whether to trigger LLM completion
  - `on_context_updated`: Optional callback triggered after context update

- Added a new foundational example `07e-interruptible-playht-http.py` for easy
  testing of `PlayHTHttpTTSService`.

- Added support for Google TTS Journey voices in `GoogleTTSService`.

- Added `29-livekit-audio-chat.py`, as a new foundational examples for
  `LiveKitTransportLayer`.

- Added `enable_prejoin_ui`, `max_participants` and `start_video_off` params
  to `DailyRoomProperties`.

- Added `session_timeout` to `FastAPIWebsocketTransport` and
  `WebsocketServerTransport` for configuring session timeouts (in
  seconds). Triggers `on_session_timeout` for custom timeout handling.
  See [examples/websocket-server/bot.py](https://github.com/pipecat-ai/pipecat/blob/main/examples/websocket-server/bot.py).

- Added the new modalities option and helper function to set Gemini output
  modalities.

- Added `examples/foundational/26d-gemini-live-text.py` which is
  using Gemini as TEXT modality and using another TTS provider for TTS process.

### Changed

- Modified `UserIdleProcessor` to start monitoring only after first
  conversation activity (`UserStartedSpeakingFrame` or
  `BotStartedSpeakingFrame`) instead of immediately.

- Modified `OpenAIAssistantContextAggregator` to support controlled completions
  and to emit context update callbacks via `FunctionCallResultProperties`.

- Added `aws_session_token` to the `PollyTTSService`.

- Changed the default model for `PlayHTHttpTTSService` to `Play3.0-mini-http`.

- `api_key`, `aws_access_key_id` and `region` are no longer required parameters
  for the PollyTTSService (AWSTTSService)

- Added `session_timeout` example in `examples/websocket-server/bot.py` to
  handle session timeout event.

- Changed `InputParams` in
  `src/pipecat/services/gemini_multimodal_live/gemini.py` to support different
  modalities.

- Changed `DeepgramSTTService` to send `finalize` event whenever VAD detects
  `UserStoppedSpeakingFrame`. This helps in faster transcriptions and clearing
  the `Deepgram` audio buffer.

### Fixed

- Fixed an issue where `DeepgramSTTService` was not generating metrics using
  pipeline's VAD.

- Fixed `UserIdleProcessor` not properly propagating `EndFrame`s through the
  pipeline.

- Fixed an issue where websocket based TTS services could incorrectly terminate
  their connection due to a retry counter not resetting.

- Fixed a `PipelineTask` issue that would cause a dangling task after stopping
  the pipeline with an `EndFrame`.

- Fixed an import issue for `PlayHTHttpTTSService`.

- Fixed an issue where languages couldn't be used with the `PlayHTHttpTTSService`.

- Fixed an issue where `OpenAIRealtimeBetaLLMService` audio chunks were hitting
  an error when truncating audio content.

- Fixed an issue where setting the voice and model for `RimeHttpTTSService`
  wasn't working.

- Fixed an issue where `IdleFrameProcessor` and `UserIdleProcessor` were getting
  initialized before the start of the pipeline.

## [0.0.52] - 2024-12-24

### Added

- Constructor arguments for GoogleLLMService to directly set tools and tool_config.

- Smart turn detection example (`22d-natural-conversation-gemini-audio.py`) that
  leverages Gemini 2.0 capabilities ().
  (see https://x.com/kwindla/status/1870974144831275410)

- Added `DailyTransport.send_dtmf()` to send dial-out DTMF tones.

- Added `DailyTransport.sip_call_transfer()` to forward SIP and PSTN calls to
  another address or number. For example, transfer a SIP call to a different
  SIP address or transfer a PSTN phone number to a different PSTN phone number.

- Added `DailyTransport.sip_refer()` to transfer incoming SIP/PSTN calls from
  outside Daily to another SIP/PSTN address.

- Added an `auto_mode` input parameter to `ElevenLabsTTSService`. `auto_mode`
  is set to `True` by default. Enabling this setting disables the chunk
  schedule and all buffers, which reduces latency.

- Added `KoalaFilter` which implement on device noise reduction using Koala
  Noise Suppression.
  (see https://picovoice.ai/platform/koala/)

- Added `CerebrasLLMService` for Cerebras integration with an OpenAI-compatible
  interface. Added foundational example `14k-function-calling-cerebras.py`.

- Pipecat now supports Python 3.13. We had a dependency on the `audioop` package
  which was deprecated and now removed on Python 3.13. We are now using
  `audioop-lts` (https://github.com/AbstractUmbra/audioop) to provide the same
  functionality.

- Added timestamped conversation transcript support:

  - New `TranscriptProcessor` factory provides access to user and assistant
    transcript processors.
  - `UserTranscriptProcessor` processes user speech with timestamps from
    transcription.
  - `AssistantTranscriptProcessor` processes assistant responses with LLM
    context timestamps.
  - Messages emitted with ISO 8601 timestamps indicating when they were spoken.
  - Supports all LLM formats (OpenAI, Anthropic, Google) via standard message
    format.
  - New examples: `28a-transcription-processor-openai.py`,
    `28b-transcription-processor-anthropic.py`, and
    `28c-transcription-processor-gemini.py`.

- Add support for more languages to ElevenLabs (Arabic, Croatian, Filipino,
  Tamil) and PlayHT (Afrikans, Albanian, Amharic, Arabic, Bengali, Croatian,
  Galician, Hebrew, Mandarin, Serbian, Tagalog, Urdu, Xhosa).

### Changed

- `PlayHTTTSService` uses the new v4 websocket API, which also fixes an issue
  where text inputted to the TTS didn't return audio.

- The default model for `ElevenLabsTTSService` is now `eleven_flash_v2_5`.

- `OpenAIRealtimeBetaLLMService` now takes a `model` parameter in the
  constructor.

- Updated the default model for the `OpenAIRealtimeBetaLLMService`.

- Room expiration (`exp`) in `DailyRoomProperties` is now optional (`None`) by
  default instead of automatically setting a 5-minute expiration time. You must
  explicitly set expiration time if desired.

### Deprecated

- `AWSTTSService` is now deprecated, use `PollyTTSService` instead.

### Fixed

- Fixed token counting in `GoogleLLMService`. Tokens were summed incorrectly
  (double-counted in many cases).

- Fixed an issue that could cause the bot to stop talking if there was a user
  interruption before getting any audio from the TTS service.

- Fixed an issue that would cause `ParallelPipeline` to handle `EndFrame`
  incorrectly causing the main pipeline to not terminate or terminate too early.

- Fixed an audio stuttering issue in `FastPitchTTSService`.

- Fixed a `BaseOutputTransport` issue that was causing non-audio frames being
  processed before the previous audio frames were played. This will allow, for
  example, sending a frame `A` after a `TTSSpeakFrame` and the frame `A` will
  only be pushed downstream after the audio generated from `TTSSpeakFrame` has
  been spoken.

- Fixed a `DeepgramSTTService` issue that was causing language to be passed as
  an object instead of a string resulting in the connection to fail.

## [0.0.51] - 2024-12-16

### Fixed

- Fixed an issue in websocket-based TTS services that was causing infinite
  reconnections (Cartesia, ElevenLabs, PlayHT and LMNT).

## [0.0.50] - 2024-12-11

### Added

- Added `GeminiMultimodalLiveLLMService`. This is an integration for Google's
  Gemini Multimodal Live API, supporting:

  - Real-time audio and video input processing
  - Streaming text responses with TTS
  - Audio transcription for both user and bot speech
  - Function calling
  - System instructions and context management
  - Dynamic parameter updates (temperature, top_p, etc.)

- Added `AudioTranscriber` utility class for handling audio transcription with
  Gemini models.

- Added new context classes for Gemini:

  - `GeminiMultimodalLiveContext`
  - `GeminiMultimodalLiveUserContextAggregator`
  - `GeminiMultimodalLiveAssistantContextAggregator`
  - `GeminiMultimodalLiveContextAggregatorPair`

- Added new foundational examples for `GeminiMultimodalLiveLLMService`:

  - `26-gemini-multimodal-live.py`
  - `26a-gemini-live-transcription.py`
  - `26b-gemini-live-video.py`
  - `26c-gemini-live-video.py`

- Added `SimliVideoService`. This is an integration for Simli AI avatars.
  (see https://www.simli.com)

- Added NVIDIA Riva's `FastPitchTTSService` and `ParakeetSTTService`.
  (see https://www.nvidia.com/en-us/ai-data-science/products/riva/)

- Added `IdentityFilter`. This is the simplest frame filter that lets through
  all incoming frames.

- New `STTMuteStrategy` called `FUNCTION_CALL` which mutes the STT service
  during LLM function calls.

- `DeepgramSTTService` now exposes two event handlers `on_speech_started` and
  `on_utterance_end` that could be used to implement interruptions. See new
  example `examples/foundational/07c-interruptible-deepgram-vad.py`.

- Added `GroqLLMService`, `GrokLLMService`, and `NimLLMService` for Groq, Grok,
  and NVIDIA NIM API integration, with an OpenAI-compatible interface.

- New examples demonstrating function calling with Groq, Grok, Azure OpenAI,
  Fireworks, and NVIDIA NIM: `14f-function-calling-groq.py`,
  `14g-function-calling-grok.py`, `14h-function-calling-azure.py`,
  `14i-function-calling-fireworks.py`, and `14j-function-calling-nvidia.py`.

- In order to obtain the audio stored by the `AudioBufferProcessor` you can now
  also register an `on_audio_data` event handler. The `on_audio_data` handler
  will be called every time `buffer_size` (a new constructor argument) is
  reached. If `buffer_size` is 0 (default) you need to manually get the audio as
  before using `AudioBufferProcessor.merge_audio_buffers()`.

```
@audiobuffer.event_handler("on_audio_data")
async def on_audio_data(processor, audio, sample_rate, num_channels):
    await save_audio(audio, sample_rate, num_channels)
```

- Added a new RTVI message called `disconnect-bot`, which when handled pushes
  an `EndFrame` to trigger the pipeline to stop.

### Changed

- `STTMuteFilter` now supports multiple simultaneous muting strategies.

- `XTTSService` language now defaults to `Language.EN`.

- `SoundfileMixer` doesn't resample input files anymore to avoid startup
  delays. The sample rate of the provided sound files now need to match the
  sample rate of the output transport.

- Input frames (audio, image and transport messages) are now system frames. This
  means they are processed immediately by all processors instead of being queued
  internally.

- Expanded the transcriptions.language module to support a superset of
  languages.

- Updated STT and TTS services with language options that match the supported
  languages for each service.

- Updated the `AzureLLMService` to use the `OpenAILLMService`. Updated the
  `api_version` to `2024-09-01-preview`.

- Updated the `FireworksLLMService` to use the `OpenAILLMService`. Updated the
  default model to `accounts/fireworks/models/firefunction-v2`.

- Updated the `simple-chatbot` example to include a Javascript and React client
  example, using RTVI JS and React.

### Removed

- Removed `AppFrame`. This was used as a special user custom frame, but there's
  actually no use case for that.

### Fixed

- Fixed a `ParallelPipeline` issue that would cause system frames to be queued.

- Fixed `FastAPIWebsocketTransport` so it can work with binary data (e.g. using
  the protobuf serializer).

- Fixed an issue in `CartesiaTTSService` that could cause previous audio to be
  received after an interruption.

- Fixed Cartesia, ElevenLabs, LMNT and PlayHT TTS websocket
  reconnection. Before, if an error occurred no reconnection was happening.

- Fixed a `BaseOutputTransport` issue that was causing audio to be discarded
  after an `EndFrame` was received.

- Fixed an issue in `WebsocketServerTransport` and `FastAPIWebsocketTransport`
  that would cause a busy loop when using audio mixer.

- Fixed a `DailyTransport` and `LiveKitTransport` issue where connections were
  being closed in the input transport prematurely. This was causing frames
  queued inside the pipeline being discarded.

- Fixed an issue in `DailyTransport` that would cause some internal callbacks to
  not be executed.

- Fixed an issue where other frames were being processed while a `CancelFrame`
  was being pushed down the pipeline.

- `AudioBufferProcessor` now handles interruptions properly.

- Fixed a `WebsocketServerTransport` issue that would prevent interruptions with
  `TwilioSerializer` from working.

- `DailyTransport.capture_participant_video` now allows capturing user's screen
  share by simply passing `video_source="screenVideo"`.

- Fixed Google Gemini message handling to properly convert appended messages to
  Gemini's required format.

- Fixed an issue with `FireworksLLMService` where chat completions were failing
  by removing the `stream_options` from the chat completion options.

## [0.0.49] - 2024-11-17

### Added

- Added RTVI `on_bot_started` event which is useful in a single turn
  interaction.

- Added `DailyTransport` events `dialin-connected`, `dialin-stopped`,
  `dialin-error` and `dialin-warning`. Needs daily-python >= 0.13.0.

- Added `RimeHttpTTSService` and the `07q-interruptible-rime.py` foundational
  example.

- Added `STTMuteFilter`, a general-purpose processor that combines STT
  muting and interruption control. When active, it prevents both transcription
  and interruptions during bot speech. The processor supports multiple
  strategies: `FIRST_SPEECH` (mute only during bot's first
  speech), `ALWAYS` (mute during all bot speech), or `CUSTOM` (using provided
  callback).

- Added `STTMuteFrame`, a control frame that enables/disables speech
  transcription in STT services.

## [0.0.48] - 2024-11-10 "Antonio release"

### Added

- There's now an input queue in each frame processor. When you call
  `FrameProcessor.push_frame()` this will internally call
  `FrameProcessor.queue_frame()` on the next processor (upstream or downstream)
  and the frame will be internally queued (except system frames). Then, the
  queued frames will get processed. With this input queue it is also possible
  for FrameProcessors to block processing more frames by calling
  `FrameProcessor.pause_processing_frames()`. The way to resume processing
  frames is by calling `FrameProcessor.resume_processing_frames()`.

- Added audio filter `NoisereduceFilter`.

- Introduce input transport audio filters (`BaseAudioFilter`). Audio filters can
  be used to remove background noises before audio is sent to VAD.

- Introduce output transport audio mixers (`BaseAudioMixer`). Output transport
  audio mixers can be used, for example, to add background sounds or any other
  audio mixing functionality before the output audio is actually written to the
  transport.

- Added `GatedOpenAILLMContextAggregator`. This aggregator keeps the last
  received OpenAI LLM context frame and it doesn't let it through until the
  notifier is notified.

- Added `WakeNotifierFilter`. This processor expects a list of frame types and
  will execute a given callback predicate when a frame of any of those type is
  being processed. If the callback returns true the notifier will be notified.

- Added `NullFilter`. A null filter doesn't push any frames upstream or
  downstream. This is usually used to disable one of the pipelines in
  `ParallelPipeline`.

- Added `EventNotifier`. This can be used as a very simple synchronization
  feature between processors.

- Added `TavusVideoService`. This is an integration for Tavus digital twins.
  (see https://www.tavus.io/)

- Added `DailyTransport.update_subscriptions()`. This allows you to have fine
  grained control of what media subscriptions you want for each participant in a
  room.

- Added audio filter `KrispFilter`.

### Changed

- The following `DailyTransport` functions are now `async` which means they need
  to be awaited: `start_dialout`, `stop_dialout`, `start_recording`,
  `stop_recording`, `capture_participant_transcription` and
  `capture_participant_video`.

- Changed default output sample rate to 24000. This changes all TTS service to
  output to 24000 and also the default output transport sample rate. This
  improves audio quality at the cost of some extra bandwidth.

- `AzureTTSService` now uses Azure websockets instead of HTTP requests.

- The previous `AzureTTSService` HTTP implementation is now
  `AzureHttpTTSService`.

### Fixed

- Websocket transports (FastAPI and Websocket) now synchronize with time before
  sending data. This allows for interruptions to just work out of the box.

- Improved bot speaking detection for all TTS services by using actual bot
  audio.

- Fixed an issue that was generating constant bot started/stopped speaking
  frames for HTTP TTS services.

- Fixed an issue that was causing stuttering with AWS TTS service.

- Fixed an issue with PlayHTTTSService, where the TTFB metrics were reporting
  very small time values.

- Fixed an issue where AzureTTSService wasn't initializing the specified
  language.

### Other

- Add `23-bot-background-sound.py` foundational example.

- Added a new foundational example `22-natural-conversation.py`. This example
  shows how to achieve a more natural conversation detecting when the user ends
  statement.

## [0.0.47] - 2024-10-22

### Added

- Added `AssemblyAISTTService` and corresponding foundational examples
  `07o-interruptible-assemblyai.py` and `13d-assemblyai-transcription.py`.

- Added a foundational example for Gladia transcription:
  `13c-gladia-transcription.py`

### Changed

- Updated `GladiaSTTService` to use the V2 API.

- Changed `DailyTransport` transcription model to `nova-2-general`.

### Fixed

- Fixed an issue that would cause an import error when importing
  `SileroVADAnalyzer` from the old package `pipecat.vad.silero`.

- Fixed `enable_usage_metrics` to control LLM/TTS usage metrics separately
  from `enable_metrics`.

## [0.0.46] - 2024-10-19

### Added

- Added `audio_passthrough` parameter to `STTService`. If enabled it allows
  audio frames to be pushed downstream in case other processors need them.

- Added input parameter options for `PlayHTTTSService` and
  `PlayHTHttpTTSService`.

### Changed

- Changed `DeepgramSTTService` model to `nova-2-general`.

- Moved `SileroVAD` audio processor to `processors.audio.vad`.

- Module `utils.audio` is now `audio.utils`. A new `resample_audio` function has
  been added.

- `PlayHTTTSService` now uses PlayHT websockets instead of HTTP requests.

- The previous `PlayHTTTSService` HTTP implementation is now
  `PlayHTHttpTTSService`.

- `PlayHTTTSService` and `PlayHTHttpTTSService` now use a `voice_engine` of
  `PlayHT3.0-mini`, which allows for multi-lingual support.

- Renamed `OpenAILLMServiceRealtimeBeta` to `OpenAIRealtimeBetaLLMService` to
  match other services.

### Deprecated

- `LLMUserResponseAggregator` and `LLMAssistantResponseAggregator` are
  mostly deprecated, use `OpenAILLMContext` instead.

- The `vad` package is now deprecated and `audio.vad` should be used
  instead. The `avd` package will get removed in a future release.

### Fixed

- Fixed an issue that would cause an error if no VAD analyzer was passed to
  `LiveKitTransport` params.

- Fixed `SileroVAD` processor to support interruptions properly.

### Other

- Added `examples/foundational/07-interruptible-vad.py`. This is the same as
  `07-interruptible.py` but using the `SileroVAD` processor instead of passing
  the `VADAnalyzer` in the transport.

## [0.0.45] - 2024-10-16

### Changed

- Metrics messages have moved out from the transport's base output into RTVI.

## [0.0.44] - 2024-10-15

### Added

- Added support for OpenAI Realtime API with the new
  `OpenAILLMServiceRealtimeBeta` processor.
  (see https://platform.openai.com/docs/guides/realtime/overview)

- Added `RTVIBotTranscriptionProcessor` which will send the RTVI
  `bot-transcription` protocol message. These are TTS text aggregated (into
  sentences) messages.

- Added new input params to the `MarkdownTextFilter` utility. You can set
  `filter_code` to filter code from text and `filter_tables` to filter tables
  from text.

- Added `CanonicalMetricsService`. This processor uses the new
  `AudioBufferProcessor` to capture conversation audio and later send it to
  Canonical AI.
  (see https://canonical.chat/)

- Added `AudioBufferProcessor`. This processor can be used to buffer mixed user and
  bot audio. This can later be saved into an audio file or processed by some
  audio analyzer.

- Added `on_first_participant_joined` event to `LiveKitTransport`.

### Changed

- LLM text responses are now logged properly as unicode characters.

- `UserStartedSpeakingFrame`, `UserStoppedSpeakingFrame`,
  `BotStartedSpeakingFrame`, `BotStoppedSpeakingFrame`, `BotSpeakingFrame` and
  `UserImageRequestFrame` are now based from `SystemFrame`

### Fixed

- Merge `RTVIBotLLMProcessor`/`RTVIBotLLMTextProcessor` and
  `RTVIBotTTSProcessor`/`RTVIBotTTSTextProcessor` to avoid out of order issues.

- Fixed an issue in RTVI protocol that could cause a `bot-llm-stopped` or
  `bot-tts-stopped` message to be sent before a `bot-llm-text` or `bot-tts-text`
  message.

- Fixed `DeepgramSTTService` constructor settings not being merged with default
  ones.

- Fixed an issue in Daily transport that would cause tasks to be hanging if
  urgent transport messages were being sent from a transport event handler.

- Fixed an issue in `BaseOutputTransport` that would cause `EndFrame` to be
  pushed downed too early and call `FrameProcessor.cleanup()` before letting the
  transport stop properly.

## [0.0.43] - 2024-10-10

### Added

- Added a new util called `MarkdownTextFilter` which is a subclass of a new
  base class called `BaseTextFilter`. This is a configurable utility which
  is intended to filter text received by TTS services.

- Added new `RTVIUserLLMTextProcessor`. This processor will send an RTVI
  `user-llm-text` message with the user content's that was sent to the LLM.

### Changed

- `TransportMessageFrame` doesn't have an `urgent` field anymore, instead
  there's now a `TransportMessageUrgentFrame` which is a `SystemFrame` and
  therefore skip all internal queuing.

- For TTS services, convert inputted languages to match each service's language
  format

### Fixed

- Fixed an issue where changing a language with the Deepgram STT service
  wouldn't apply the change. This was fixed by disconnecting and reconnecting
  when the language changes.

## [0.0.42] - 2024-10-02

### Added

- `SentryMetrics` has been added to report frame processor metrics to
  Sentry. This is now possible because `FrameProcessorMetrics` can now be passed
  to `FrameProcessor`.

- Added Google TTS service and corresponding foundational example
  `07n-interruptible-google.py`

- Added AWS Polly TTS support and `07m-interruptible-aws.py` as an example.

- Added InputParams to Azure TTS service.

- Added `LivekitTransport` (audio-only for now).

- RTVI 0.2.0 is now supported.

- All `FrameProcessors` can now register event handlers.

```
tts = SomeTTSService(...)

@tts.event_handler("on_connected"):
async def on_connected(processor):
  ...
```

- Added `AsyncGeneratorProcessor`. This processor can be used together with a
  `FrameSerializer` as an async generator. It provides a `generator()` function
  that returns an `AsyncGenerator` and that yields serialized frames.

- Added `EndTaskFrame` and `CancelTaskFrame`. These are new frames that are
  meant to be pushed upstream to tell the pipeline task to stop nicely or
  immediately respectively.

- Added configurable LLM parameters (e.g., temperature, top_p, max_tokens, seed)
  for OpenAI, Anthropic, and Together AI services along with corresponding
  setter functions.

- Added `sample_rate` as a constructor parameter for TTS services.

- Pipecat has a pipeline-based architecture. The pipeline consists of frame
  processors linked to each other. The elements traveling across the pipeline
  are called frames.

  To have a deterministic behavior the frames traveling through the pipeline
  should always be ordered, except system frames which are out-of-band
  frames. To achieve that, each frame processor should only output frames from a
  single task.

  In this version all the frame processors have their own task to push
  frames. That is, when `push_frame()` is called the given frame will be put
  into an internal queue (with the exception of system frames) and a frame
  processor task will push it out.

- Added pipeline clocks. A pipeline clock is used by the output transport to
  know when a frame needs to be presented. For that, all frames now have an
  optional `pts` field (prensentation timestamp). There's currently just one
  clock implementation `SystemClock` and the `pts` field is currently only used
  for `TextFrame`s (audio and image frames will be next).

- A clock can now be specified to `PipelineTask` (defaults to
  `SystemClock`). This clock will be passed to each frame processor via the
  `StartFrame`.

- Added `CartesiaHttpTTSService`.

- `DailyTransport` now supports setting the audio bitrate to improve audio
  quality through the `DailyParams.audio_out_bitrate` parameter. The new
  default is 96kbps.

- `DailyTransport` now uses the number of audio output channels (1 or 2) to set
  mono or stereo audio when needed.

- Interruptions support has been added to `TwilioFrameSerializer` when using
  `FastAPIWebsocketTransport`.

- Added new `LmntTTSService` text-to-speech service.
  (see https://www.lmnt.com/)

- Added `TTSModelUpdateFrame`, `TTSLanguageUpdateFrame`, `STTModelUpdateFrame`,
  and `STTLanguageUpdateFrame` frames to allow you to switch models, language
  and voices in TTS and STT services.

- Added new `transcriptions.Language` enum.

### Changed

- Context frames are now pushed downstream from assistant context aggregators.

- Removed Silero VAD torch dependency.

- Updated individual update settings frame classes into a single
  `ServiceUpdateSettingsFrame` class.

- We now distinguish between input and output audio and image frames. We
  introduce `InputAudioRawFrame`, `OutputAudioRawFrame`, `InputImageRawFrame`
  and `OutputImageRawFrame` (and other subclasses of those). The input frames
  usually come from an input transport and are meant to be processed inside the
  pipeline to generate new frames. However, the input frames will not be sent
  through an output transport. The output frames can also be processed by any
  frame processor in the pipeline and they are allowed to be sent by the output
  transport.

- `ParallelTask` has been renamed to `SyncParallelPipeline`. A
  `SyncParallelPipeline` is a frame processor that contains a list of different
  pipelines to be executed concurrently. The difference between a
  `SyncParallelPipeline` and a `ParallelPipeline` is that, given an input frame,
  the `SyncParallelPipeline` will wait for all the internal pipelines to
  complete. This is achieved by making sure the last processor in each of the
  pipelines is synchronous (e.g. an HTTP-based service that waits for the
  response).

- `StartFrame` is back a system frame to make sure it's processed immediately by
  all processors. `EndFrame` stays a control frame since it needs to be ordered
  allowing the frames in the pipeline to be processed.

- Updated `MoondreamService` revision to `2024-08-26`.

- `CartesiaTTSService` and `ElevenLabsTTSService` now add presentation
  timestamps to their text output. This allows the output transport to push the
  text frames downstream at almost the same time the words are spoken. We say
  "almost" because currently the audio frames don't have presentation timestamp
  but they should be played at roughly the same time.

- `DailyTransport.on_joined` event now returns the full session data instead of
  just the participant.

- `CartesiaTTSService` is now a subclass of `TTSService`.

- `DeepgramSTTService` is now a subclass of `STTService`.

- `WhisperSTTService` is now a subclass of `SegmentedSTTService`. A
  `SegmentedSTTService` is a `STTService` where the provided audio is given in a
  big chunk (i.e. from when the user starts speaking until the user stops
  speaking) instead of a continous stream.

### Fixed

- Fixed OpenAI multiple function calls.

- Fixed a Cartesia TTS issue that would cause audio to be truncated in some
  cases.

- Fixed a `BaseOutputTransport` issue that would stop audio and video rendering
  tasks (after receiving and `EndFrame`) before the internal queue was emptied,
  causing the pipeline to finish prematurely.

- `StartFrame` should be the first frame every processor receives to avoid
  situations where things are not initialized (because initialization happens on
  `StartFrame`) and other frames come in resulting in undesired behavior.

### Performance

- `obj_id()` and `obj_count()` now use `itertools.count` avoiding the need of
  `threading.Lock`.

### Other

- Pipecat now uses Ruff as its formatter (https://github.com/astral-sh/ruff).

## [0.0.41] - 2024-08-22

### Added

- Added `LivekitFrameSerializer` audio frame serializer.

### Fixed

- Fix `FastAPIWebsocketOutputTransport` variable name clash with subclass.

- Fix an `AnthropicLLMService` issue with empty arguments in function calling.

### Other

- Fixed `studypal` example errors.

## [0.0.40] - 2024-08-20

### Added

- VAD parameters can now be dynamicallt updated using the
  `VADParamsUpdateFrame`.

- `ErrorFrame` has now a `fatal` field to indicate the bot should exit if a
  fatal error is pushed upstream (false by default). A new `FatalErrorFrame`
  that sets this flag to true has been added.

- `AnthropicLLMService` now supports function calling and initial support for
  prompt caching.
  (see https://www.anthropic.com/news/prompt-caching)

- `ElevenLabsTTSService` can now specify ElevenLabs input parameters such as
  `output_format`.

- `TwilioFrameSerializer` can now specify Twilio's and Pipecat's desired sample
  rates to use.

- Added new `on_participant_updated` event to `DailyTransport`.

- Added `DailyRESTHelper.delete_room_by_name()` and
  `DailyRESTHelper.delete_room_by_url()`.

- Added LLM and TTS usage metrics. Those are enabled when
  `PipelineParams.enable_usage_metrics` is True.

- `AudioRawFrame`s are now pushed downstream from the base output
  transport. This allows capturing the exact words the bot says by adding an STT
  service at the end of the pipeline.

- Added new `GStreamerPipelineSource`. This processor can generate image or
  audio frames from a GStreamer pipeline (e.g. reading an MP4 file, and RTP
  stream or anything supported by GStreamer).

- Added `TransportParams.audio_out_is_live`. This flag is False by default and
  it is useful to indicate we should not synchronize audio with sporadic images.

- Added new `BotStartedSpeakingFrame` and `BotStoppedSpeakingFrame` control
  frames. These frames are pushed upstream and they should wrap
  `BotSpeakingFrame`.

- Transports now allow you to register event handlers without decorators.

### Changed

- Support RTVI message protocol 0.1. This includes new messages, support for
  messages responses, support for actions, configuration, webhooks and a bunch
  of new cool stuff.
  (see https://docs.rtvi.ai/)

- `SileroVAD` dependency is now imported via pip's `silero-vad` package.

- `ElevenLabsTTSService` now uses `eleven_turbo_v2_5` model by default.

- `BotSpeakingFrame` is now a control frame.

- `StartFrame` is now a control frame similar to `EndFrame`.

- `DeepgramTTSService` now is more customizable. You can adjust the encoding and
  sample rate.

### Fixed

- `TTSStartFrame` and `TTSStopFrame` are now sent when TTS really starts and
  stops. This allows for knowing when the bot starts and stops speaking even
  with asynchronous services (like Cartesia).

- Fixed `AzureSTTService` transcription frame timestamps.

- Fixed an issue with `DailyRESTHelper.create_room()` expirations which would
  cause this function to stop working after the initial expiration elapsed.

- Improved `EndFrame` and `CancelFrame` handling. `EndFrame` should end things
  gracefully while a `CancelFrame` should cancel all running tasks as soon as
  possible.

- Fixed an issue in `AIService` that would cause a yielded `None` value to be
  processed.

- RTVI's `bot-ready` message is now sent when the RTVI pipeline is ready and
  a first participant joins.

- Fixed a `BaseInputTransport` issue that was causing incoming system frames to
  be queued instead of being pushed immediately.

- Fixed a `BaseInputTransport` issue that was causing start/stop interruptions
  incoming frames to not cancel tasks and be processed properly.

### Other

- Added `studypal` example (from to the Cartesia folks!).

- Most examples now use Cartesia.

- Added examples `foundational/19a-tools-anthropic.py`,
  `foundational/19b-tools-video-anthropic.py` and
  `foundational/19a-tools-togetherai.py`.

- Added examples `foundational/18-gstreamer-filesrc.py` and
  `foundational/18a-gstreamer-videotestsrc.py` that show how to use
  `GStreamerPipelineSource`

- Remove `requests` library usage.

- Cleanup examples and use `DailyRESTHelper`.

## [0.0.39] - 2024-07-23

### Fixed

- Fixed a regression introduced in 0.0.38 that would cause Daily transcription
  to stop the Pipeline.

## [0.0.38] - 2024-07-23

### Added

- Added `force_reload`, `skip_validation` and `trust_repo` to `SileroVAD` and
  `SileroVADAnalyzer`. This allows caching and various GitHub repo validations.

- Added `send_initial_empty_metrics` flag to `PipelineParams` to request for
  initial empty metrics (zero values). True by default.

### Fixed

- Fixed initial metrics format. It was using the wrong keys name/time instead of
  processor/value.

- STT services should be using ISO 8601 time format for transcription frames.

- Fixed an issue that would cause Daily transport to show a stop transcription
  error when actually none occurred.

## [0.0.37] - 2024-07-22

### Added

- Added `RTVIProcessor` which implements the RTVI-AI standard.
  See https://github.com/rtvi-ai

- Added `BotInterruptionFrame` which allows interrupting the bot while talking.

- Added `LLMMessagesAppendFrame` which allows appending messages to the current
  LLM context.

- Added `LLMMessagesUpdateFrame` which allows changing the LLM context for the
  one provided in this new frame.

- Added `LLMModelUpdateFrame` which allows updating the LLM model.

- Added `TTSSpeakFrame` which causes the bot say some text. This text will not
  be part of the LLM context.

- Added `TTSVoiceUpdateFrame` which allows updating the TTS voice.

### Removed

- We remove the `LLMResponseStartFrame` and `LLMResponseEndFrame` frames. These
  were added in the past to properly handle interruptions for the
  `LLMAssistantContextAggregator`. But the `LLMContextAggregator` is now based
  on `LLMResponseAggregator` which handles interruptions properly by just
  processing the `StartInterruptionFrame`, so there's no need for these extra
  frames any more.

### Fixed

- Fixed an issue with `StatelessTextTransformer` where it was pushing a string
  instead of a `TextFrame`.

- `TTSService` end of sentence detection has been improved. It now works with
  acronyms, numbers, hours and others.

- Fixed an issue in `TTSService` that would not properly flush the current
  aggregated sentence if an `LLMFullResponseEndFrame` was found.

### Performance

- `CartesiaTTSService` now uses websockets which improves speed. It also
  leverages the new Cartesia contexts which maintains generated audio prosody
  when multiple inputs are sent, therefore improving audio quality a lot.

## [0.0.36] - 2024-07-02

### Added

- Added `GladiaSTTService`.
  See https://docs.gladia.io/chapters/speech-to-text-api/pages/live-speech-recognition

- Added `XTTSService`. This is a local Text-To-Speech service.
  See https://github.com/coqui-ai/TTS

- Added `UserIdleProcessor`. This processor can be used to wait for any
  interaction with the user. If the user doesn't say anything within a given
  timeout a provided callback is called.

- Added `IdleFrameProcessor`. This processor can be used to wait for frames
  within a given timeout. If no frame is received within the timeout a provided
  callback is called.

- Added new frame `BotSpeakingFrame`. This frame will be continuously pushed
  upstream while the bot is talking.

- It is now possible to specify a Silero VAD version when using `SileroVADAnalyzer`
  or `SileroVAD`.

- Added `AysncFrameProcessor` and `AsyncAIService`. Some services like
  `DeepgramSTTService` need to process things asynchronously. For example, audio
  is sent to Deepgram but transcriptions are not returned immediately. In these
  cases we still require all frames (except system frames) to be pushed
  downstream from a single task. That's what `AsyncFrameProcessor` is for. It
  creates a task and all frames should be pushed from that task. So, whenever a
  new Deepgram transcription is ready that transcription will also be pushed
  from this internal task.

- The `MetricsFrame` now includes processing metrics if metrics are enabled. The
  processing metrics indicate the time a processor needs to generate all its
  output. Note that not all processors generate these kind of metrics.

### Changed

- `WhisperSTTService` model can now also be a string.

- Added missing \* keyword separators in services.

### Fixed

- `WebsocketServerTransport` doesn't try to send frames anymore if serializers
  returns `None`.

- Fixed an issue where exceptions that occurred inside frame processors were
  being swallowed and not displayed.

- Fixed an issue in `FastAPIWebsocketTransport` where it would still try to send
  data to the websocket after being closed.

### Other

- Added Fly.io deployment example in `examples/deployment/flyio-example`.

- Added new `17-detect-user-idle.py` example that shows how to use the new
  `UserIdleProcessor`.

## [0.0.35] - 2024-06-28

### Changed

- `FastAPIWebsocketParams` now require a serializer.

- `TwilioFrameSerializer` now requires a `streamSid`.

### Fixed

- Silero VAD number of frames needs to be 512 for 16000 sample rate or 256 for
  8000 sample rate.

## [0.0.34] - 2024-06-25

### Fixed

- Fixed an issue with asynchronous STT services (Deepgram and Azure) that could
  interruptions to ignore transcriptions.

- Fixed an issue introduced in 0.0.33 that would cause the LLM to generate
  shorter output.

## [0.0.33] - 2024-06-25

### Changed

- Upgraded to Cartesia's new Python library 1.0.0. `CartesiaTTSService` now
  expects a voice ID instead of a voice name (you can get the voice ID from
  Cartesia's playground). You can also specify the audio `sample_rate` and
  `encoding` instead of the previous `output_format`.

### Fixed

- Fixed an issue with asynchronous STT services (Deepgram and Azure) that could
  cause static audio issues and interruptions to not work properly when dealing
  with multiple LLMs sentences.

- Fixed an issue that could mix new LLM responses with previous ones when
  handling interruptions.

- Fixed a Daily transport blocking situation that occurred while reading audio
  frames after a participant left the room. Needs daily-python >= 0.10.1.

## [0.0.32] - 2024-06-22

### Added

- Allow specifying a `DeepgramSTTService` url which allows using on-prem
  Deepgram.

- Added new `FastAPIWebsocketTransport`. This is a new websocket transport that
  can be integrated with FastAPI websockets.

- Added new `TwilioFrameSerializer`. This is a new serializer that knows how to
  serialize and deserialize audio frames from Twilio.

- Added Daily transport event: `on_dialout_answered`. See
  https://reference-python.daily.co/api_reference.html#daily.EventHandler

- Added new `AzureSTTService`. This allows you to use Azure Speech-To-Text.

### Performance

- Convert `BaseOutputTransport` and `BaseOutputTransport` to fully use asyncio
  and remove the use of threads.

### Other

- Added `twilio-chatbot`. This is an example that shows how to integrate Twilio
  phone numbers with a Pipecat bot.

- Updated `07f-interruptible-azure.py` to use `AzureLLMService`,
  `AzureSTTService` and `AzureTTSService`.

## [0.0.31] - 2024-06-13

### Performance

- Break long audio frames into 20ms chunks instead of 10ms.

## [0.0.30] - 2024-06-13

### Added

- Added `report_only_initial_ttfb` to `PipelineParams`. This will make it so
  only the initial TTFB metrics after the user stops talking are reported.

- Added `OpenPipeLLMService`. This service will let you run OpenAI through
  OpenPipe's SDK.

- Allow specifying frame processors' name through a new `name` constructor
  argument.

- Added `DeepgramSTTService`. This service has an ongoing websocket
  connection. To handle this, it subclasses `AIService` instead of
  `STTService`. The output of this service will be pushed from the same task,
  except system frames like `StartFrame`, `CancelFrame` or
  `StartInterruptionFrame`.

### Changed

- `FrameSerializer.deserialize()` can now return `None` in case it is not
  possible to desearialize the given data.

- `daily_rest.DailyRoomProperties` now allows extra unknown parameters.

### Fixed

- Fixed an issue where `DailyRoomProperties.exp` always had the same old
  timestamp unless set by the user.

- Fixed a couple of issues with `WebsocketServerTransport`. It needed to use
  `push_audio_frame()` and also VAD was not working properly.

- Fixed an issue that would cause LLM aggregator to fail with small
  `VADParams.stop_secs` values.

- Fixed an issue where `BaseOutputTransport` would send longer audio frames
  preventing interruptions.

### Other

- Added new `07h-interruptible-openpipe.py` example. This example shows how to
  use OpenPipe to run OpenAI LLMs and get the logs stored in OpenPipe.

- Added new `dialin-chatbot` example. This examples shows how to call the bot
  using a phone number.

## [0.0.29] - 2024-06-07

### Added

- Added a new `FunctionFilter`. This filter will let you filter frames based on
  a given function, except system messages which should never be filtered.

- Added `FrameProcessor.can_generate_metrics()` method to indicate if a
  processor can generate metrics. In the future this might get an extra argument
  to ask for a specific type of metric.

- Added `BasePipeline`. All pipeline classes should be based on this class. All
  subclasses should implement a `processors_with_metrics()` method that returns
  a list of all `FrameProcessor`s in the pipeline that can generate metrics.

- Added `enable_metrics` to `PipelineParams`.

- Added `MetricsFrame`. The `MetricsFrame` will report different metrics in the
  system. Right now, it can report TTFB (Time To First Byte) values for
  different services, that is the time spent between the arrival of a `Frame` to
  the processor/service until the first `DataFrame` is pushed downstream. If
  metrics are enabled an intial `MetricsFrame` with all the services in the
  pipeline will be sent.

- Added TTFB metrics and debug logging for TTS services.

### Changed

- Moved `ParallelTask` to `pipecat.pipeline.parallel_task`.

### Fixed

- Fixed PlayHT TTS service to work properly async.

## [0.0.28] - 2024-06-05

### Fixed

- Fixed an issue with `SileroVADAnalyzer` that would cause memory to keep
  growing indefinitely.

## [0.0.27] - 2024-06-05

### Added

- Added `DailyTransport.participants()` and `DailyTransport.participant_counts()`.

## [0.0.26] - 2024-06-05

### Added

- Added `OpenAITTSService`.

- Allow passing `output_format` and `model_id` to `CartesiaTTSService` to change
  audio sample format and the model to use.

- Added `DailyRESTHelper` which helps you create Daily rooms and tokens in an
  easy way.

- `PipelineTask` now has a `has_finished()` method to indicate if the task has
  completed. If a task is never ran `has_finished()` will return False.

- `PipelineRunner` now supports SIGTERM. If received, the runner will be
  cancelled.

### Fixed

- Fixed an issue where `BaseInputTransport` and `BaseOutputTransport` where
  stopping push tasks before pushing `EndFrame` frames could cause the bots to
  get stuck.

- Fixed an error closing local audio transports.

- Fixed an issue with Deepgram TTS that was introduced in the previous release.

- Fixed `AnthropicLLMService` interruptions. If an interruption occurred, a
  `user` message could be appended after the previous `user` message. Anthropic
  does not allow that because it requires alternate `user` and `assistant`
  messages.

### Performance

- The `BaseInputTransport` does not pull audio frames from sub-classes any
  more. Instead, sub-classes now push audio frames into a queue in the base
  class. Also, `DailyInputTransport` now pushes audio frames every 20ms instead
  of 10ms.

- Remove redundant camera input thread from `DailyInputTransport`. This should
  improve performance a little bit when processing participant videos.

- Load Cartesia voice on startup.

## [0.0.25] - 2024-05-31

### Added

- Added WebsocketServerTransport. This will create a websocket server and will
  read messages coming from a client. The messages are serialized/deserialized
  with protobufs. See `examples/websocket-server` for a detailed example.

- Added function calling (LLMService.register_function()). This will allow the
  LLM to call functions you have registered when needed. For example, if you
  register a function to get the weather in Los Angeles and ask the LLM about
  the weather in Los Angeles, the LLM will call your function.
  See https://platform.openai.com/docs/guides/function-calling

- Added new `LangchainProcessor`.

- Added Cartesia TTS support (https://cartesia.ai/)

### Fixed

- Fixed SileroVAD frame processor.

- Fixed an issue where `camera_out_enabled` would cause the highg CPU usage if
  no image was provided.

### Performance

- Removed unnecessary audio input tasks.

## [0.0.24] - 2024-05-29

### Added

- Exposed `on_dialin_ready` for Daily transport SIP endpoint handling. This
  notifies when the Daily room SIP endpoints are ready. This allows integrating
  with third-party services like Twilio.

- Exposed Daily transport `on_app_message` event.

- Added Daily transport `on_call_state_updated` event.

- Added Daily transport `start_recording()`, `stop_recording` and
  `stop_dialout`.

### Changed

- Added `PipelineParams`. This replaces the `allow_interruptions` argument in
  `PipelineTask` and will allow future parameters in the future.

- Fixed Deepgram Aura TTS base_url and added ErrorFrame reporting.

- GoogleLLMService `api_key` argument is now mandatory.

### Fixed

- Daily tranport `dialin-ready` doesn't not block anymore and it now handles
  timeouts.

- Fixed AzureLLMService.

## [0.0.23] - 2024-05-23

### Fixed

- Fixed an issue handling Daily transport `dialin-ready` event.

## [0.0.22] - 2024-05-23

### Added

- Added Daily transport `start_dialout()` to be able to make phone or SIP calls.
  See https://reference-python.daily.co/api_reference.html#daily.CallClient.start_dialout

- Added Daily transport support for dial-in use cases.

- Added Daily transport events: `on_dialout_connected`, `on_dialout_stopped`,
  `on_dialout_error` and `on_dialout_warning`. See
  https://reference-python.daily.co/api_reference.html#daily.EventHandler

## [0.0.21] - 2024-05-22

### Added

- Added vision support to Anthropic service.

- Added `WakeCheckFilter` which allows you to pass information downstream only
  if you say a certain phrase/word.

### Changed

- `FrameSerializer.serialize()` and `FrameSerializer.deserialize()` are now
  `async`.

- `Filter` has been renamed to `FrameFilter` and it's now under
  `processors/filters`.

### Fixed

- Fixed Anthropic service to use new frame types.

- Fixed an issue in `LLMUserResponseAggregator` and `UserResponseAggregator`
  that would cause frames after a brief pause to not be pushed to the LLM.

- Clear the audio output buffer if we are interrupted.

- Re-add exponential smoothing after volume calculation. This makes sure the
  volume value being used doesn't fluctuate so much.

## [0.0.20] - 2024-05-22

### Added

- In order to improve interruptions we now compute a loudness level using
  [pyloudnorm](https://github.com/csteinmetz1/pyloudnorm). The audio coming
  WebRTC transports (e.g. Daily) have an Automatic Gain Control (AGC) algorithm
  applied to the signal, however we don't do that on our local PyAudio
  signals. This means that currently incoming audio from PyAudio is kind of
  broken. We will fix it in future releases.

### Fixed

- Fixed an issue where `StartInterruptionFrame` would cause
  `LLMUserResponseAggregator` to push the accumulated text causing the LLM
  respond in the wrong task. The `StartInterruptionFrame` should not trigger any
  new LLM response because that would be spoken in a different task.

- Fixed an issue where tasks and threads could be paused because the executor
  didn't have more tasks available. This was causing issues when cancelling and
  recreating tasks during interruptions.

## [0.0.19] - 2024-05-20

### Changed

- `LLMUserResponseAggregator` and `LLMAssistantResponseAggregator` internal
  messages are now exposed through the `messages` property.

### Fixed

- Fixed an issue where `LLMAssistantResponseAggregator` was not accumulating the
  full response but short sentences instead. If there's an interruption we only
  accumulate what the bot has spoken until now in a long response as well.

## [0.0.18] - 2024-05-20

### Fixed

- Fixed an issue in `DailyOuputTransport` where transport messages were not
  being sent.

## [0.0.17] - 2024-05-19

### Added

- Added `google.generativeai` model support, including vision. This new `google`
  service defaults to using `gemini-1.5-flash-latest`. Example in
  `examples/foundational/12a-describe-video-gemini-flash.py`.

- Added vision support to `openai` service. Example in
  `examples/foundational/12a-describe-video-gemini-flash.py`.

- Added initial interruptions support. The assistant contexts (or aggregators)
  should now be placed after the output transport. This way, only the completed
  spoken context is added to the assistant context.

- Added `VADParams` so you can control voice confidence level and others.

- `VADAnalyzer` now uses an exponential smoothed volume to improve speech
  detection. This is useful when voice confidence is high (because there's
  someone talking near you) but volume is low.

### Fixed

- Fixed an issue where TTSService was not pushing TextFrames downstream.

- Fixed issues with Ctrl-C program termination.

- Fixed an issue that was causing `StopTaskFrame` to actually not exit the
  `PipelineTask`.

## [0.0.16] - 2024-05-16

### Fixed

- `DailyTransport`: don't publish camera and audio tracks if not enabled.

- Fixed an issue in `BaseInputTransport` that was causing frames pushed
  downstream not pushed in the right order.

## [0.0.15] - 2024-05-15

### Fixed

- Quick hot fix for receiving `DailyTransportMessage`.

## [0.0.14] - 2024-05-15

### Added

- Added `DailyTransport` event `on_participant_left`.

- Added support for receiving `DailyTransportMessage`.

### Fixed

- Images are now resized to the size of the output camera. This was causing
  images not being displayed.

- Fixed an issue in `DailyTransport` that would not allow the input processor to
  shutdown if no participant ever joined the room.

- Fixed base transports start and stop. In some situation processors would halt
  or not shutdown properly.

## [0.0.13] - 2024-05-14

### Changed

- `MoondreamService` argument `model_id` is now `model`.

- `VADAnalyzer` arguments have been renamed for more clarity.

### Fixed

- Fixed an issue with `DailyInputTransport` and `DailyOutputTransport` that
  could cause some threads to not start properly.

- Fixed `STTService`. Add `max_silence_secs` and `max_buffer_secs` to handle
  better what's being passed to the STT service. Also add exponential smoothing
  to the RMS.

- Fixed `WhisperSTTService`. Add `no_speech_prob` to avoid garbage output text.

## [0.0.12] - 2024-05-14

### Added

- Added `DailyTranscriptionSettings` to be able to specify transcription
  settings much easier (e.g. language).

### Other

- Updated `simple-chatbot` with Spanish.

- Add missing dependencies in some of the examples.

## [0.0.11] - 2024-05-13

### Added

- Allow stopping pipeline tasks with new `StopTaskFrame`.

### Changed

- TTS, STT and image generation service now use `AsyncGenerator`.

### Fixed

- `DailyTransport`: allow registering for participant transcriptions even if
  input transport is not initialized yet.

### Other

- Updated `storytelling-chatbot`.

## [0.0.10] - 2024-05-13

### Added

- Added Intel GPU support to `MoondreamService`.

- Added support for sending transport messages (e.g. to communicate with an app
  at the other end of the transport).

- Added `FrameProcessor.push_error()` to easily send an `ErrorFrame` upstream.

### Fixed

- Fixed Azure services (TTS and image generation).

### Other

- Updated `simple-chatbot`, `moondream-chatbot` and `translation-chatbot`
  examples.

## [0.0.9] - 2024-05-12

### Changed

Many things have changed in this version. Many of the main ideas such as frames,
processors, services and transports are still there but some things have changed
a bit.

- `Frame`s describe the basic units for processing. For example, text, image or
  audio frames. Or control frames to indicate a user has started or stopped
  speaking.

- `FrameProcessor`s process frames (e.g. they convert a `TextFrame` to an
  `ImageRawFrame`) and push new frames downstream or upstream to their linked
  peers.

- `FrameProcessor`s can be linked together. The easiest wait is to use the
  `Pipeline` which is a container for processors. Linking processors allow
  frames to travel upstream or downstream easily.

- `Transport`s are a way to send or receive frames. There can be local
  transports (e.g. local audio or native apps), network transports
  (e.g. websocket) or service transports (e.g. https://daily.co).

- `Pipeline`s are just a processor container for other processors.

- A `PipelineTask` know how to run a pipeline.

- A `PipelineRunner` can run one or more tasks and it is also used, for example,
  to capture Ctrl-C from the user.

## [0.0.8] - 2024-04-11

### Added

- Added `FireworksLLMService`.

- Added `InterimTranscriptionFrame` and enable interim results in
  `DailyTransport` transcriptions.

### Changed

- `FalImageGenService` now uses new `fal_client` package.

### Fixed

- `FalImageGenService`: use `asyncio.to_thread` to not block main loop when
  generating images.

- Allow `TranscriptionFrame` after an end frame (transcriptions can be delayed
  and received after `UserStoppedSpeakingFrame`).

## [0.0.7] - 2024-04-10

### Added

- Add `use_cpu` argument to `MoondreamService`.

## [0.0.6] - 2024-04-10

### Added

- Added `FalImageGenService.InputParams`.

- Added `URLImageFrame` and `UserImageFrame`.

- Added `UserImageRequestFrame` and allow requesting an image from a participant.

- Added base `VisionService` and `MoondreamService`

### Changed

- Don't pass `image_size` to `ImageGenService`, images should have their own size.

- `ImageFrame` now receives a tuple`(width,height)` to specify the size.

- `on_first_other_participant_joined` now gets a participant argument.

### Fixed

- Check if camera, speaker and microphone are enabled before writing to them.

### Performance

- `DailyTransport` only subscribe to desired participant video track.

## [0.0.5] - 2024-04-06

### Changed

- Use `camera_bitrate` and `camera_framerate`.

- Increase `camera_framerate` to 30 by default.

### Fixed

- Fixed `LocalTransport.read_audio_frames`.

## [0.0.4] - 2024-04-04

### Added

- Added project optional dependencies `[silero,openai,...]`.

### Changed

- Moved thransports to its own directory.

- Use `OPENAI_API_KEY` instead of `OPENAI_CHATGPT_API_KEY`.

### Fixed

- Don't write to microphone/speaker if not enabled.

### Other

- Added live translation example.

- Fix foundational examples.

## [0.0.3] - 2024-03-13

### Other

- Added `storybot` and `chatbot` examples.

## [0.0.2] - 2024-03-12

Initial public release.