> ## Documentation Index > Fetch the complete documentation index at: https://developers.telnyx.com/llms.txt > Use this file to discover all available pages before exploring further. # Stream text to speech over WebSocket > Open a WebSocket connection to stream text and receive synthesized audio in real time. Authentication is provided via the standard `Authorization: Bearer ` header. Send JSON frames with text to synthesize; receive JSON frames containing base64-encoded audio chunks. Supported providers: `aws`, `telnyx`, `azure`, `murfai`, `minimax`, `rime`, `resemble`, `elevenlabs`, `xai`. **Connection flow:** 1. Open WebSocket with query parameters specifying provider, voice, and model. 2. Send an initial handshake message `{"text": " "}` (single space) with optional `voice_settings` to initialize the session. 3. Send text messages as `{"text": "Hello world"}`. 4. Receive audio chunks as JSON frames with base64-encoded audio. 5. A final frame with `isFinal: true` indicates the end of audio for the current text. To interrupt and restart synthesis mid-stream, send `{"force": true}` — the current worker is stopped and a new one is started. **Note:** The Telnyx `Ultra` model is not available over WebSocket. Use the HTTP POST `/text-to-speech/speech` endpoint instead. ## OpenAPI ````yaml https://telnyx-openapi-ng.s3.us-east-1.amazonaws.com/text-to-speech/text-to-speech.yml get /text-to-speech openapi: 3.1.0 info: title: Text to Speech API version: 2.0.0 description: API for managing Text to Speech. contact: email: support@telnyx.com servers: - url: https://api.telnyx.com/v2 security: - bearerAuth: [] tags: - name: Text to Speech description: Text to Speech operations paths: /text-to-speech: get: tags: - Text To Speech Commands summary: Stream text to speech over WebSocket description: >- Open a WebSocket connection to stream text and receive synthesized audio in real time. Authentication is provided via the standard `Authorization: Bearer ` header. Send JSON frames with text to synthesize; receive JSON frames containing base64-encoded audio chunks. Supported providers: `aws`, `telnyx`, `azure`, `murfai`, `minimax`, `rime`, `resemble`, `elevenlabs`, `xai`. **Connection flow:** 1. Open WebSocket with query parameters specifying provider, voice, and model. 2. Send an initial handshake message `{"text": " "}` (single space) with optional `voice_settings` to initialize the session. 3. Send text messages as `{"text": "Hello world"}`. 4. Receive audio chunks as JSON frames with base64-encoded audio. 5. A final frame with `isFinal: true` indicates the end of audio for the current text. To interrupt and restart synthesis mid-stream, send `{"force": true}` — the current worker is stopped and a new one is started. **Note:** The Telnyx `Ultra` model is not available over WebSocket. Use the HTTP POST `/text-to-speech/speech` endpoint instead. operationId: TextToSpeechOverWs parameters: - $ref: '#/components/parameters/voice' - $ref: '#/components/parameters/provider' - $ref: '#/components/parameters/model_id' - $ref: '#/components/parameters/voice_id' - $ref: '#/components/parameters/disable_cache' - $ref: '#/components/parameters/audio_format' - $ref: '#/components/parameters/socket_id' responses: '101': description: >- WebSocket connection established. Communication proceeds via JSON frames. **Client → Server:** See `ClientTextFrame` schema. **Server → Client:** See `AudioChunkFrame`, `FinalFrame`, and `ErrorFrame` schemas. content: application/json: schema: oneOf: - $ref: '#/components/schemas/ClientTextFrame' - $ref: '#/components/schemas/TtsServerEvent' '200': description: >- WebSocket upgrade successful — this response is not returned directly. See 101 for frame documentation. '400': description: >- Invalid parameters — provider not supported or missing required fields. '401': description: >- Authentication failed — missing or invalid `x-telnyx-auth-rev2` header. components: parameters: voice: name: voice in: query required: false description: >- Voice identifier in the format `provider.model_id.voice_id` or `provider.voice_id` (e.g. `telnyx.NaturalHD.Telnyx_Alloy`, `Telnyx.Ultra.`, or `azure.en-US-AvaMultilingualNeural`). When provided, the `provider`, `model_id`, and `voice_id` are extracted automatically. Takes precedence over individual `provider`/`model_id`/`voice_id` parameters. schema: type: string provider: name: provider in: query required: false description: >- TTS provider. Defaults to `telnyx` if not specified. Ignored when `voice` is provided. schema: type: string enum: - aws - telnyx - azure - elevenlabs - minimax - murfai - rime - resemble - xai default: telnyx model_id: name: model_id in: query required: false description: >- Model identifier for the chosen provider. Examples: `Natural`, `NaturalHD`, `Ultra` (Telnyx); `Polly.Generative` (AWS). schema: type: string voice_id: name: voice_id in: query required: false description: Voice identifier for the chosen provider. schema: type: string disable_cache: name: disable_cache in: query required: false description: When `true`, bypass the audio cache and generate fresh audio. schema: type: boolean default: false audio_format: name: audio_format in: query required: false description: >- Audio output format override. Supported for Telnyx models. `pcm` and `wav` are available for `Natural`/`NaturalHD` models. The `Ultra` model outputs PCM at 24kHz s16le or MP3 at 128kbps 24kHz. schema: type: string enum: - pcm - wav - mp3 socket_id: name: socket_id in: query required: false description: >- Client-provided socket identifier for tracking. If not provided, one is generated server-side. schema: type: string schemas: ClientTextFrame: type: object description: Client-to-server frame containing text to synthesize. required: - text properties: text: type: string description: >- Text to convert to speech. Send `" "` (single space) as an initial handshake with optional `voice_settings`. Subsequent messages contain the actual text to synthesize. voice_settings: type: object description: >- Provider-specific voice settings sent with the initial handshake. Contents vary by provider — e.g. `{"speed": 1.2}` for Minimax, `{"voice_speed": 1.5}` for Telnyx. additionalProperties: true force: type: boolean description: >- When `true`, stops the current synthesis worker and starts a new one. Used to interrupt speech mid-stream and begin synthesizing new text. examples: - text: ' ' voice_settings: voice_speed: 1.2 - text: Hello, welcome to Telnyx text to speech. - text: New text after interruption. force: true TtsServerEvent: description: Union of all server-to-client WebSocket events for TTS streaming. oneOf: - $ref: '#/components/schemas/AudioChunkFrame' - $ref: '#/components/schemas/FinalFrame' - $ref: '#/components/schemas/ErrorFrame' discriminator: propertyName: type mapping: audio_chunk: $ref: '#/components/schemas/AudioChunkFrame' final: $ref: '#/components/schemas/FinalFrame' error: $ref: '#/components/schemas/ErrorFrame' AudioChunkFrame: type: object description: Server-to-client frame containing a base64-encoded audio chunk. properties: type: type: string const: audio_chunk description: Frame type identifier. audio: type: - string - 'null' description: >- Base64-encoded audio data. May be `null` for providers that use `drop_concatenated_audio` mode (Telnyx Natural/NaturalHD, Rime, Minimax, MurfAI, Resemble) — in that case only streamed chunks carry audio. text: type: - string - 'null' description: The text segment that this audio chunk corresponds to. isFinal: type: boolean description: Always `false` for audio chunk frames. cached: type: boolean description: Whether this audio was served from cache. timeToFirstAudioFrameMs: type: integer description: >- Milliseconds from the start-of-speech request to the first audio frame. Only present on the first audio chunk of a synthesis request. FinalFrame: type: object description: >- Server-to-client frame indicating synthesis is complete for the current text. properties: type: type: string const: final description: Frame type identifier. audio: type: 'null' description: Always `null` for the final frame. text: type: string description: Empty string. isFinal: type: boolean description: Always `true`. const: true timeToFirstAudioFrameMs: type: integer description: Present if this was the first response frame. ErrorFrame: type: object description: >- Server-to-client frame indicating an error during synthesis. The connection is closed shortly after. properties: type: type: string const: error description: Frame type identifier. error: type: string description: Error message describing what went wrong. securitySchemes: bearerAuth: scheme: bearer type: http ````