Cartesia Streaming WebSocket APIs

asyncapi: 2.6.0
info:
  title: Cartesia Streaming WebSocket APIs
  version: 2026-03-01
  description: |
    AsyncAPI description of Cartesia's real-time WebSocket interfaces, derived
    strictly from the official Cartesia documentation at https://docs.cartesia.ai.

    Three WebSocket surfaces are covered:

    - Sonic TTS WebSocket (`/tts/websocket`) — text-to-speech streaming with
      contexts, continuations, flushes, word and phoneme timestamps.
    - Ink STT WebSocket (`/stt/websocket`) — streaming speech-to-text with
      delta transcripts and manual finalize/close control frames.
    - Ink STT Turns WebSocket (`/stt/turns/websocket`) — streaming
      speech-to-text with native turn detection events.

    All connections require the `Cartesia-Version` header or `cartesia_version`
    query parameter, and authenticate via an `X-API-Key` header or an
    `access_token` query parameter (the latter is the only browser-safe option
    because browser WebSocket APIs cannot set custom headers).
  contact:
    name: Cartesia Documentation
    url: https://docs.cartesia.ai
  license:
    name: Cartesia Terms of Service
    url: https://cartesia.ai/legal/terms-of-service
  termsOfService: https://cartesia.ai/legal/terms-of-service
  tags:
    - name: TTS
    - name: STT
    - name: Streaming
    - name: WebSocket
    - name: Sonic
    - name: Ink

defaultContentType: application/json

servers:
  tts:
    url: api.cartesia.ai
    protocol: wss
    description: Sonic text-to-speech streaming WebSocket server.
    bindings:
      ws:
        bindingVersion: 0.1.0
        headers:
          type: object
          properties:
            Cartesia-Version:
              type: string
              description: API version date string (e.g. 2026-03-01). Required unless supplied as the `cartesia_version` query parameter.
              example: 2026-03-01
            X-API-Key:
              type: string
              description: Cartesia API key. Required unless using `access_token` query parameter.
        query:
          type: object
          properties:
            cartesia_version:
              type: string
              description: API version (alternative to the Cartesia-Version header; required for browser clients).
              example: 2026-03-01
            access_token:
              type: string
              description: Short-lived access token for browser clients that cannot set custom headers.
    security:
      - apiKeyHeader: []
      - accessTokenQuery: []
  stt:
    url: api.cartesia.ai
    protocol: wss
    description: Ink streaming speech-to-text WebSocket server (manual finalize).
    bindings:
      ws:
        bindingVersion: 0.1.0
        headers:
          type: object
          properties:
            X-API-Key:
              type: string
        query:
          type: object
          required:
            - model
            - encoding
            - sample_rate
            - cartesia_version
          properties:
            model:
              type: string
              description: STT model identifier (e.g. ink-2, ink-whisper).
              example: ink-2
            encoding:
              type: string
              enum:
                - pcm_s16le
                - pcm_s32le
                - pcm_f16le
                - pcm_f32le
                - pcm_mulaw
                - pcm_alaw
            sample_rate:
              type: integer
              description: Audio sample rate in Hz.
            cartesia_version:
              type: string
              example: 2026-03-01
            language:
              type: string
              description: ISO-639-1 language code. Defaults to `en`.
              default: en
            min_volume:
              type: number
              minimum: 0
              maximum: 1
              description: Silence detection threshold (ink-whisper only).
            max_silence_duration_secs:
              type: number
              description: Auto-finalize threshold in seconds (ink-whisper only).
            access_token:
              type: string
              description: Short-lived access token for browser clients.
    security:
      - apiKeyHeader: []
      - accessTokenQuery: []
  stt_turns:
    url: api.cartesia.ai
    protocol: wss
    description: Ink streaming speech-to-text WebSocket server with native turn detection.
    bindings:
      ws:
        bindingVersion: 0.1.0
        headers:
          type: object
          properties:
            X-API-Key:
              type: string
        query:
          type: object
          required:
            - model
            - encoding
            - sample_rate
            - cartesia_version
          properties:
            model:
              type: string
              example: ink-2
            encoding:
              type: string
              enum:
                - pcm_s16le
                - pcm_s32le
                - pcm_f16le
                - pcm_f32le
                - pcm_mulaw
                - pcm_alaw
            sample_rate:
              type: integer
            cartesia_version:
              type: string
              example: 2026-03-01
            access_token:
              type: string
    security:
      - apiKeyHeader: []
      - accessTokenQuery: []

channels:
  /tts/websocket:
    servers:
      - tts
    description: |
      Sonic TTS bidirectional WebSocket. Clients send JSON `generationRequest`
      or `cancelRequest` frames; the server returns JSON `chunk`,
      `flush_done`, `done`, `timestamps`, `phoneme_timestamps`, and `error`
      frames. Multiple concurrent contexts are multiplexed by `context_id`.
    bindings:
      ws:
        bindingVersion: 0.1.0
        method: GET
    publish:
      operationId: ttsClientFrame
      summary: Client → Server frames for Sonic TTS.
      message:
        oneOf:
          - $ref: '#/components/messages/TtsGenerationRequest'
          - $ref: '#/components/messages/TtsCancelRequest'
    subscribe:
      operationId: ttsServerFrame
      summary: Server → Client frames for Sonic TTS.
      message:
        oneOf:
          - $ref: '#/components/messages/TtsChunk'
          - $ref: '#/components/messages/TtsTimestamps'
          - $ref: '#/components/messages/TtsPhonemeTimestamps'
          - $ref: '#/components/messages/TtsFlushDone'
          - $ref: '#/components/messages/TtsDone'
          - $ref: '#/components/messages/TtsError'

  /stt/websocket:
    servers:
      - stt
    description: |
      Ink STT streaming WebSocket. Clients send binary audio frames matching
      the encoding/sample_rate query parameters, plus optional textual
      `finalize` and `close` control frames. The server emits JSON
      `transcript`, `flush_done`, `done`, and `error` frames.
    bindings:
      ws:
        bindingVersion: 0.1.0
        method: GET
    publish:
      operationId: sttClientFrame
      summary: Client → Server frames for Ink STT.
      message:
        oneOf:
          - $ref: '#/components/messages/SttAudioBinary'
          - $ref: '#/components/messages/SttFinalize'
          - $ref: '#/components/messages/SttClose'
    subscribe:
      operationId: sttServerFrame
      summary: Server → Client frames for Ink STT.
      message:
        oneOf:
          - $ref: '#/components/messages/SttTranscript'
          - $ref: '#/components/messages/SttFlushDone'
          - $ref: '#/components/messages/SttDone'
          - $ref: '#/components/messages/SttError'

  /stt/turns/websocket:
    servers:
      - stt_turns
    description: |
      Ink STT Turns WebSocket. Clients send binary audio frames plus an
      optional JSON `close` control frame. The server emits `connected`,
      `turn.start`, `turn.update`, `turn.eager_end`, `turn.resume`,
      `turn.end`, and `error` events.
    bindings:
      ws:
        bindingVersion: 0.1.0
        method: GET
    publish:
      operationId: sttTurnsClientFrame
      summary: Client → Server frames for Ink STT Turns.
      message:
        oneOf:
          - $ref: '#/components/messages/SttAudioBinary'
          - $ref: '#/components/messages/SttTurnsClose'
    subscribe:
      operationId: sttTurnsServerFrame
      summary: Server → Client frames for Ink STT Turns.
      message:
        oneOf:
          - $ref: '#/components/messages/SttTurnsConnected'
          - $ref: '#/components/messages/SttTurnStart'
          - $ref: '#/components/messages/SttTurnUpdate'
          - $ref: '#/components/messages/SttTurnEagerEnd'
          - $ref: '#/components/messages/SttTurnResume'
          - $ref: '#/components/messages/SttTurnEnd'
          - $ref: '#/components/messages/SttTurnsError'

components:
  securitySchemes:
    apiKeyHeader:
      type: httpApiKey
      name: X-API-Key
      in: header
      description: Cartesia API key.
    accessTokenQuery:
      type: httpApiKey
      name: access_token
      in: query
      description: Short-lived access token suitable for browser clients that cannot set custom headers.

  messages:
    # ----- TTS client frames -----
    TtsGenerationRequest:
      name: ttsGenerationRequest
      title: TTS Generation Request
      summary: Submit (or continue) a transcript for streaming synthesis on a context.
      contentType: application/json
      payload:
        $ref: '#/components/schemas/TtsGenerationRequest'
    TtsCancelRequest:
      name: ttsCancelRequest
      title: TTS Cancel Request
      summary: Terminate an in-flight generation for a context.
      contentType: application/json
      payload:
        $ref: '#/components/schemas/TtsCancelRequest'

    # ----- TTS server frames -----
    TtsChunk:
      name: ttsChunk
      title: TTS Audio Chunk
      summary: A base64-encoded audio chunk for an active context.
      contentType: application/json
      payload:
        $ref: '#/components/schemas/TtsChunk'
    TtsTimestamps:
      name: ttsTimestamps
      title: TTS Word Timestamps
      summary: Word-level start/end timings for synthesized audio.
      contentType: application/json
      payload:
        $ref: '#/components/schemas/TtsTimestamps'
    TtsPhonemeTimestamps:
      name: ttsPhonemeTimestamps
      title: TTS Phoneme Timestamps
      summary: Phoneme-level start/end timings for synthesized audio.
      contentType: application/json
      payload:
        $ref: '#/components/schemas/TtsPhonemeTimestamps'
    TtsFlushDone:
      name: ttsFlushDone
      title: TTS Flush Done
      summary: Acknowledgement that a flush boundary has been emitted on the context.
      contentType: application/json
      payload:
        $ref: '#/components/schemas/TtsFlushDone'
    TtsDone:
      name: ttsDone
      title: TTS Done
      summary: Final generation-complete signal for a context.
      contentType: application/json
      payload:
        $ref: '#/components/schemas/TtsDone'
    TtsError:
      name: ttsError
      title: TTS Error
      summary: Error condition on the TTS WebSocket or a specific context.
      contentType: application/json
      payload:
        $ref: '#/components/schemas/TtsError'

    # ----- STT client frames -----
    SttAudioBinary:
      name: sttAudioBinary
      title: STT Audio Frame
      summary: Raw audio bytes matching the negotiated encoding and sample rate.
      contentType: application/octet-stream
      payload:
        type: string
        format: binary
        description: Binary WebSocket frame carrying ~100ms of audio.
    SttFinalize:
      name: sttFinalize
      title: STT Finalize Control
      summary: Plain-text `finalize` control frame; triggers transcription of buffered audio.
      contentType: text/plain
      payload:
        type: string
        const: finalize
    SttClose:
      name: sttClose
      title: STT Close Control
      summary: Plain-text `close` control frame; flushes remaining audio and ends the session.
      contentType: text/plain
      payload:
        type: string
        const: close

    # ----- STT server frames -----
    SttTranscript:
      name: sttTranscript
      title: STT Transcript
      summary: Delta transcript with word-level timing.
      contentType: application/json
      payload:
        $ref: '#/components/schemas/SttTranscript'
    SttFlushDone:
      name: sttFlushDone
      title: STT Flush Done
      summary: Acknowledgement of a `finalize` control frame.
      contentType: application/json
      payload:
        $ref: '#/components/schemas/SttFlushDone'
    SttDone:
      name: sttDone
      title: STT Done
      summary: Acknowledgement of a `close` control frame.
      contentType: application/json
      payload:
        $ref: '#/components/schemas/SttDone'
    SttError:
      name: sttError
      title: STT Error
      summary: Error condition on the STT WebSocket.
      contentType: application/json
      payload:
        $ref: '#/components/schemas/SttError'

    # ----- STT Turns frames -----
    SttTurnsClose:
      name: sttTurnsClose
      title: STT Turns Close
      summary: 'JSON `{ "type": "close" }` control frame for the Turns WebSocket.'
      contentType: application/json
      payload:
        type: object
        required: [type]
        properties:
          type:
            type: string
            const: close
    SttTurnsConnected:
      name: sttTurnsConnected
      title: STT Turns Connected
      contentType: application/json
      payload:
        $ref: '#/components/schemas/SttTurnsConnected'
    SttTurnStart:
      name: sttTurnStart
      title: STT Turn Start
      contentType: application/json
      payload:
        $ref: '#/components/schemas/SttTurnStart'
    SttTurnUpdate:
      name: sttTurnUpdate
      title: STT Turn Update
      contentType: application/json
      payload:
        $ref: '#/components/schemas/SttTurnUpdate'
    SttTurnEagerEnd:
      name: sttTurnEagerEnd
      title: STT Turn Eager End
      contentType: application/json
      payload:
        $ref: '#/components/schemas/SttTurnEagerEnd'
    SttTurnResume:
      name: sttTurnResume
      title: STT Turn Resume
      contentType: application/json
      payload:
        $ref: '#/components/schemas/SttTurnResume'
    SttTurnEnd:
      name: sttTurnEnd
      title: STT Turn End
      contentType: application/json
      payload:
        $ref: '#/components/schemas/SttTurnEnd'
    SttTurnsError:
      name: sttTurnsError
      title: STT Turns Error
      contentType: application/json
      payload:
        $ref: '#/components/schemas/SttError'

  schemas:
    # ===== TTS schemas =====
    TtsVoice:
      type: object
      required: [mode, id]
      properties:
        mode:
          type: string
          enum: [id]
          description: Voice selection mode. Only `id` is supported on the WebSocket.
        id:
          type: string
          description: Voice UUID from the Cartesia voice library or a cloned voice.

    TtsOutputFormat:
      type: object
      required: [container, encoding, sample_rate]
      properties:
        container:
          type: string
          enum: [raw]
        encoding:
          type: string
          enum:
            - pcm_f32le
            - pcm_s16le
            - pcm_mulaw
            - pcm_alaw
        sample_rate:
          type: integer
          enum: [8000, 16000, 22050, 24000, 44100, 48000]

    TtsEmotion:
      type: string
      description: Emotion preset for sonic-3 and sonic-3.5.
      enum:
        - Happy
        - Excited
        - Enthusiastic
        - Elated
        - Euphoric
        - Triumphant
        - Amazed
        - Surprised
        - Flirtatious
        - Joking/Comedic
        - Curious
        - Content
        - Peaceful
        - Serene
        - Calm
        - Grateful
        - Affectionate
        - Trust
        - Sympathetic
        - Anticipation
        - Mysterious
        - Angry
        - Mad
        - Outraged
        - Frustrated
        - Agitated
        - Threatened
        - Disgusted
        - Contempt
        - Envious
        - Sarcastic
        - Ironic
        - Sad
        - Dejected
        - Melancholic
        - Disappointed
        - Hurt
        - Guilty
        - Bored
        - Tired
        - Rejected
        - Nostalgic
        - Wistful
        - Apologetic
        - Hesitant
        - Insecure
        - Confused
        - Resigned
        - Anxious
        - Panicked
        - Alarmed
        - Scared
        - Neutral
        - Proud
        - Confident
        - Distant
        - Skeptical
        - Contemplative
        - Determined

    TtsGenerationConfig:
      type: object
      description: Speech attribute configuration for sonic-3 and sonic-3.5.
      properties:
        volume:
          type: number
          minimum: 0.5
          maximum: 2.0
          default: 1.0
        speed:
          type: number
          minimum: 0.6
          maximum: 1.5
          default: 1.0
        emotion:
          $ref: '#/components/schemas/TtsEmotion'

    TtsLanguage:
      type: string
      enum:
        - en
        - fr
        - de
        - es
        - pt
        - zh
        - ja
        - hi
        - it
        - ko
        - nl
        - pl
        - ru
        - sv
        - tr
        - tl
        - bg
        - ro
        - ar
        - cs
        - el
        - fi
        - hr
        - ms
        - sk
        - da
        - ta
        - uk
        - hu
        - "no"
        - vi
        - bn
        - th
        - he
        - ka
        - id
        - te
        - gu
        - kn
        - ml
        - mr
        - pa

    TtsGenerationRequest:
      type: object
      required:
        - model_id
        - transcript
        - voice
        - output_format
        - context_id
      properties:
        model_id:
          type: string
          description: Sonic model ID, e.g. `sonic-3.5`, `sonic-3.5-2026-05-04`, `sonic-latest`, `sonic-3`, `sonic-2`, `sonic-turbo`, `sonic`.
          example: sonic-3.5
        transcript:
          type: string
          description: Text chunk to synthesize on this context.
        voice:
          $ref: '#/components/schemas/TtsVoice'
        output_format:
          $ref: '#/components/schemas/TtsOutputFormat'
        context_id:
          type: string
          description: Unique identifier for the generation context. Reuse to continue/flush.
        language:
          $ref: '#/components/schemas/TtsLanguage'
        continue:
          type: boolean
          default: false
          description: When true, indicates more transcript chunks will follow on this context.
        max_buffer_delay_ms:
          type: integer
          minimum: 0
          maximum: 5000
          default: 3000
          description: Maximum buffer delay in milliseconds before emitting audio.
        flush:
          type: boolean
          description: When true, forces a flush boundary on the context.
        add_timestamps:
          type: boolean
          default: false
          description: Emit word-level `timestamps` frames.
        add_phoneme_timestamps:
          type: boolean
          default: false
          description: Emit `phoneme_timestamps` frames.
        use_normalized_timestamps:
          type: boolean
          description: Use normalized (vs. original) text timestamps.
        pronunciation_dict_id:
          type: string
          description: Pronunciation dictionary ID (sonic-3 and newer).
        generation_config:
          $ref: '#/components/schemas/TtsGenerationConfig'
        speed:
          type: string
          enum: [slow, normal, fast]
          deprecated: true
          description: Deprecated speed preset; use `generation_config.speed` instead.

    TtsCancelRequest:
      type: object
      required:
        - context_id
        - cancel
      properties:
        context_id:
          type: string
        cancel:
          type: boolean
          const: true

    TtsChunk:
      type: object
      required: [type, data, done, status_code, step_time, context_id]
      properties:
        type:
          type: string
          const: chunk
        data:
          type: string
          format: byte
          description: Base64-encoded audio bytes in the negotiated encoding.
        done:
          type: boolean
          description: True when this is the final chunk for the context.
        status_code:
          type: integer
          description: HTTP-style status code.
        step_time:
          type: number
          description: Server processing time for the step in milliseconds.
        context_id:
          type: string

    TtsTimestamps:
      type: object
      required: [type, done, status_code, context_id]
      properties:
        type:
          type: string
          const: timestamps
        done:
          type: boolean
        status_code:
          type: integer
        context_id:
          type: string
        word_timestamps:
          type: object
          properties:
            words:
              type: array
              items:
                type: string
            start:
              type: array
              items:
                type: number
            end:
              type: array
              items:
                type: number

    TtsPhonemeTimestamps:
      type: object
      required: [type, done, status_code, context_id]
      properties:
        type:
          type: string
          const: phoneme_timestamps
        done:
          type: boolean
        status_code:
          type: integer
        context_id:
          type: string
        phoneme_timestamps:
          type: object
          properties:
            phonemes:
              type: array
              items:
                type: string
            start:
              type: array
              items:
                type: number
            end:
              type: array
              items:
                type: number

    TtsFlushDone:
      type: object
      required: [type, done, flush_done, flush_id, status_code, context_id]
      properties:
        type:
          type: string
          const: flush_done
        done:
          type: boolean
        flush_done:
          type: boolean
          const: true
        flush_id:
          type: integer
          minimum: 1
          description: Monotonic counter of flushes on this context (starts at 1).
        status_code:
          type: integer
        context_id:
          type: string

    TtsDone:
      type: object
      required: [type, done, status_code, context_id]
      properties:
        type:
          type: string
          const: done
        done:
          type: boolean
          const: true
        status_code:
          type: integer
        context_id:
          type: string

    TtsError:
      type: object
      required: [type, done]
      properties:
        type:
          type: string
          const: error
        done:
          type: boolean
        error_code:
          type: string
          description: Machine-readable error code.
        status_code:
          type: number
          description: HTTP-style status code.
        title:
          type: string
        message:
          type: string
        doc_url:
          type: string
          format: uri
        request_id:
          type: string
          description: WebSocket connection identifier.
        context_id:
          type: string

    # ===== STT schemas =====
    SttWord:
      type: object
      required: [word, start, end]
      properties:
        word:
          type: string
        start:
          type: number
          description: Start time in seconds relative to stream start.
        end:
          type: number
          description: End time in seconds relative to stream start.

    SttTranscript:
      type: object
      required: [type, is_final, request_id, text]
      properties:
        type:
          type: string
          const: transcript
        is_final:
          type: boolean
          description: True when this delta is the final version for the segment.
        request_id:
          type: string
          format: uuid
        text:
          type: string
          description: Delta from the last finalized chunk.
        duration:
          type: number
        words:
          type: array
          items:
            $ref: '#/components/schemas/SttWord'

    SttFlushDone:
      type: object
      required: [type, request_id]
      properties:
        type:
          type: string
          const: flush_done
        request_id:
          type: string
          format: uuid

    SttDone:
      type: object
      required: [type, request_id]
      properties:
        type:
          type: string
          const: done
        request_id:
          type: string
          format: uuid

    SttError:
      type: object
      required: [type, status_code, title, message]
      properties:
        type:
          type: string
          const: error
        status_code:
          type: number
        title:
          type: string
        message:
          type: string
        error_code:
          type: string
          nullable: true
        doc_url:
          type: string
          format: uri
        request_id:
          type: string
          format: uuid

    # ===== STT Turns schemas =====
    SttTurnsConnected:
      type: object
      required: [type, request_id]
      properties:
        type:
          type: string
          const: connected
        request_id:
          type: string
          format: uuid

    SttTurnStart:
      type: object
      required: [type, request_id]
      properties:
        type:
          type: string
          const: turn.start
        request_id:
          type: string
          format: uuid

    SttTurnUpdate:
      type: object
      required: [type, transcript, request_id]
      properties:
        type:
          type: string
          const: turn.update
        transcript:
          type: string
          description: Cumulative transcription for the in-progress turn.
        request_id:
          type: string
          format: uuid

    SttTurnEagerEnd:
      type: object
      required: [type, transcript, request_id]
      properties:
        type:
          type: string
          const: turn.eager_end
        transcript:
          type: string
        request_id:
          type: string
          format: uuid

    SttTurnResume:
      type: object
      required: [type, request_id]
      properties:
        type:
          type: string
          const: turn.resume
        request_id:
          type: string
          format: uuid

    SttTurnEnd:
      type: object
      required: [type, transcript, request_id]
      properties:
        type:
          type: string
          const: turn.end
        transcript:
          type: string
        request_id:
          type: string
          format: uuid