vocabulary:
  - term: text-to-speech
    abbreviation: TTS
    definition: >
      The process of converting written text into spoken audio. Neuphonic's TTS
      engine delivers sub-25ms latency for the first audio chunk.

  - term: Server-Sent Events
    abbreviation: SSE
    definition: >
      A web technology allowing a server to push real-time updates to a client
      over a single HTTP connection. Neuphonic uses SSE to stream PCM audio
      chunks from the /sse/speak/{lang_code} endpoint.

  - term: WebSocket
    definition: >
      A full-duplex communication protocol over a persistent TCP connection.
      Neuphonic's WebSocket TTS endpoint (wss://api.neuphonic.com) enables
      continuous, low-latency voice streaming with a 60-second connection timeout.

  - term: voice cloning
    definition: >
      The process of creating a synthetic voice model from audio samples. Neuphonic
      requires a minimum of 6 seconds of audio (MP3 or WAV, max 10MB) and supports
      cloning in English, French, German, and Spanish.

  - term: lang_code
    definition: >
      A two-letter language code identifying the language for TTS synthesis or
      voice cloning. Neuphonic supports: en (English), es (Spanish), de (German),
      fr (French), ur (Urdu), ja (Japanese), ko (Korean), zh (Chinese), pt (Portuguese).

  - term: PCM
    abbreviation: Pulse Code Modulation
    definition: >
      A raw, uncompressed digital audio format. Neuphonic SSE streams return
      audio in PCM format, allowing developers to control sampling rate
      (e.g., 22050 Hz or 44100 Hz).

  - term: STOP token
    definition: >
      A special signal sent over a Neuphonic WebSocket connection to indicate
      that no more text will be sent and the server should finalize the audio
      stream.

  - term: sampling_rate
    definition: >
      The number of audio samples captured per second, measured in Hz. Neuphonic
      supports configurable sampling rates including 8000, 16000, 22050, 44100,
      and 48000 Hz.

  - term: agent
    definition: >
      A Neuphonic conversational AI voice agent that combines Neuphonic TTS with
      GPT-4o for interactive voice applications. Agents support MCP server
      integration, interruption handling, and multilingual operation.

  - term: Model Context Protocol
    abbreviation: MCP
    definition: >
      An open standard for connecting AI models to external tools and data sources.
      Neuphonic agents can integrate with MCP servers to extend their capabilities
      (e.g., retrieving order data, querying databases).

  - term: LiveKit
    definition: >
      An open-source WebRTC infrastructure platform used by Neuphonic for real-time
      agent communication sessions. Clients obtain a LiveKit token via
      POST /agents/{agent_id}/token to join an agent session.

  - term: JWT
    abbreviation: JSON Web Token
    definition: >
      A compact, URL-safe token format used for authentication. Neuphonic's
      /sse/auth endpoint exchanges an API key for a short-lived JWT to enable
      faster authentication on subsequent SSE requests.

  - term: X-API-KEY
    definition: >
      The HTTP request header used to pass a Neuphonic API key for authenticating
      SSE and REST API requests. WebSocket connections instead pass the key as an
      api_key query parameter.

  - term: NeuTTS Air
    definition: >
      Neuphonic's open-source on-device TTS model designed to run without a GPU.
      Suitable for edge deployments where cloud connectivity is limited or
      privacy is paramount.

  - term: NeuTTS Nano
    definition: >
      Neuphonic's compact open-source on-device TTS model optimized for minimal
      resource usage. Runs on-device without a GPU.

  - term: latency
    definition: >
      The time delay between sending a TTS request and receiving the first audio
      chunk. Neuphonic targets sub-25ms latency for the initial audio response
      in streaming mode.