vocabulary:
  - term: pre-recorded
    definition: >
      Audio submitted for asynchronous transcription after the recording has completed.
      Jobs are queued and results fetched by polling the job ID endpoint.

  - term: live transcription
    definition: >
      Real-time speech-to-text conversion of a streaming audio session via WebSocket.
      Individual sessions are capped at 3 hours; latency is sub-second.

  - term: diarization
    definition: >
      Speaker separation: the process of partitioning an audio stream into segments
      attributed to individual speakers. Gladia supports automatic or fixed speaker count.

  - term: audio intelligence
    definition: >
      LLM-powered enrichments applied to a transcript, including summarization,
      sentiment analysis, topic detection, named-entity recognition, and custom prompts.

  - term: x-gladia-key
    definition: >
      The HTTP request header used to authenticate all Gladia API calls. The value is
      the API key obtained from the Gladia dashboard.

  - term: transcription job
    definition: >
      An asynchronous processing unit identified by a UUID. Callers submit audio,
      receive a job ID, and poll GET /v2/pre-recorded/{id} for status and results.

  - term: code switching
    definition: >
      A multilingual mode in which the model detects and transcribes speech that
      alternates between two or more languages within a single utterance or session.

  - term: utterance
    definition: >
      A single continuous segment of speech attributed to one speaker, with associated
      start time, end time, confidence score, and text content.

  - term: word-level timestamps
    definition: >
      Per-word start and end times within a transcript, enabling precise alignment of
      text to the audio waveform.

  - term: endpointing
    definition: >
      In live sessions, the silence duration (in milliseconds) after which the model
      commits a completed utterance and emits a final transcript segment.

  - term: audio upload
    definition: >
      The process of providing audio to Gladia either by uploading a file via
      multipart/form-data to POST /v2/upload or by supplying an accessible audio_url.

  - term: callback
    definition: >
      An optional webhook URL to which Gladia POSTs a result payload when an
      asynchronous transcription job completes, eliminating the need for polling.

  - term: speaker diarization config
    definition: >
      Configuration for diarization including number_of_speakers (exact count),
      min_speakers, and max_speakers to bound automatic speaker detection.

  - term: translation
    definition: >
      Post-transcription machine translation of the transcript into one or more
      target languages, applied as an audio intelligence enrichment step.

  - term: PII redaction
    definition: >
      Automatic detection and removal or replacement of personally identifiable
      information (names, phone numbers, email addresses, etc.) in the transcript.

  - term: subtitles
    definition: >
      Export of the transcript formatted as subtitle captions (SRT or VTT) with
      configurable style and maximum words per caption segment.

  - term: summarization
    definition: >
      LLM-generated summary of the transcript, available as a bullet-point or
      paragraph format audio intelligence enrichment.

  - term: models endpoint
    definition: >
      GET /v1/models returns Gladia's available transcription models in the
      OpenRouter integration spec format.

  - term: language config
    definition: >
      Object specifying which languages to expect in the audio and whether
      code-switching between languages should be enabled.

  - term: context prompt
    definition: >
      A short natural-language description of the audio content passed to the
      model to improve recognition of domain-specific terminology.