vocabulary:
  name: LMNT API Vocabulary
  description: Domain terms and concepts for the LMNT Speech API platform.
  url: https://docs.lmnt.com/
  version: "1.0"
  created: "2026-06-12"

terms:

  - term: Voice
    definition: >
      A voice model available for text-to-speech synthesis. Voices can be system-provided
      (intrinsic) or user-created through voice cloning. Each voice has a unique ID, lifecycle
      state, ownership scope, and optional metadata such as gender, description, and tags.
    related:
      - Voice Cloning
      - Synthesis

  - term: Voice Cloning
    definition: >
      The process of creating a custom voice from audio samples. LMNT supports instant cloning
      from as little as 5 seconds of audio. Professional cloning produces higher-quality results
      from more training data.
    related:
      - Voice
      - Instant Clone
      - Professional Clone

  - term: Instant Clone
    definition: >
      A rapid voice cloning type that can generate a custom voice from minimal audio input
      (as little as 5 seconds). Trades some quality for speed of creation.
    related:
      - Voice Cloning
      - Professional Clone

  - term: Professional Clone
    definition: >
      A high-quality voice cloning type that requires more audio input but produces a more
      accurate and expressive representation of the target voice.
    related:
      - Voice Cloning
      - Instant Clone

  - term: Synthesis
    definition: >
      The process of converting text input into spoken audio using a selected voice model.
      LMNT synthesis supports streaming binary output, word timestamps, and multiple audio formats.
    related:
      - Voice
      - Speech Session
      - Text-to-Speech

  - term: Speech Session
    definition: >
      A WebSocket-based real-time streaming connection for text-to-speech generation optimized
      for LLM pipeline integration. Speech sessions support reset-latency handling for
      conversational AI interrupt scenarios.
    related:
      - Synthesis
      - WebSocket
      - Streaming

  - term: Blizzard 2
    definition: >
      LMNT's production text-to-speech model optimized for accuracy, expressiveness, and
      pronunciation. The model achieves sub-300ms latency for real-time conversational AI use cases.
    related:
      - Synthesis
      - Model

  - term: Sample Rate
    definition: >
      The number of audio samples per second in the synthesized output, measured in Hz.
      LMNT supports 8000 Hz, 16000 Hz, and 24000 Hz sample rates.
    related:
      - Synthesis
      - Audio Format

  - term: Temperature
    definition: >
      A parameter controlling the expressiveness of synthesized speech. Higher temperature
      values produce more varied and expressive output; lower values produce more consistent,
      monotone delivery.
    related:
      - Synthesis
      - Speed

  - term: Speed
    definition: >
      A multiplier applied to the speaking rate of synthesized speech. A value of 1.0 is
      normal speed; values below 1.0 slow speech down and values above 1.0 speed it up.
      Typical range is 0.25 to 2.0.
    related:
      - Synthesis
      - Temperature

  - term: Word Timestamps
    definition: >
      Optional metadata returned alongside synthesized audio indicating the start and end
      time of each word in the output. Useful for caption generation and lip-sync applications.
    related:
      - Synthesis

  - term: Streaming Audio
    definition: >
      Audio output delivered progressively as it is generated, rather than waiting for the
      full synthesis to complete. LMNT supports streaming binary audio for both REST and
      WebSocket endpoints.
    related:
      - Speech Session
      - Synthesis

  - term: API Key
    definition: >
      The authentication credential used to access LMNT APIs. API keys are managed via the
      LMNT app dashboard and passed in the X-API-Key request header.
    related:
      - Authentication

  - term: Reset Latency
    definition: >
      A feature of LMNT Speech Sessions that allows an in-progress synthesis stream to be
      cancelled and restarted quickly, enabling natural interrupt handling in conversational
      AI applications.
    related:
      - Speech Session
      - Streaming Audio