> For clean Markdown of any page, append .md to the page URL. > For a complete documentation index, see https://docs.narilabs.com/llms.txt. > For AI client integration (Claude Code, Cursor, etc.), connect to the MCP server at https://docs.narilabs.com/_mcp/server. # Stream text to speech GET /v1/text-to-speech/{voice_id}/stream-input Stream text and receive base64-encoded 24 kHz PCM audio over an authenticated WebSocket. ## Message flow 1. Send `{"text":" "}` immediately after connecting. This initialization space is not billed. 2. Send text fragments while receiving `audio` chunks. Set `flush:true` to finish a segment without closing the connection. Audio may arrive before a flush. 3. Send `{"text":""}` to end input. Keep receiving until `{"isFinal":true}`, then the connection closes. Output streams automatically. Only automatic generation is supported. There is no public flush-completion event; `isFinal` completes the whole connection. See the [TTS input streaming guide](/streaming-text) for a runnable example. Reference: https://docs.narilabs.com/api-reference/text-to-speech/input-stream/stream-text-to-speech ## AsyncAPI Specification ```yaml asyncapi: 2.6.0 info: title: Stream text to speech version: subpackage_inputStream.Stream text to speech description: > Stream text and receive base64-encoded 24 kHz PCM audio over an authenticated WebSocket. ## Message flow 1. Send `{"text":" "}` immediately after connecting. This initialization space is not billed. 2. Send text fragments while receiving `audio` chunks. Set `flush:true` to finish a segment without closing the connection. Audio may arrive before a flush. 3. Send `{"text":""}` to end input. Keep receiving until `{"isFinal":true}`, then the connection closes. Output streams automatically. Only automatic generation is supported. There is no public flush-completion event; `isFinal` completes the whole connection. See the [TTS input streaming guide](/streaming-text) for a runnable example. channels: /v1/text-to-speech/{voice_id}/stream-input: description: > Stream text and receive base64-encoded 24 kHz PCM audio over an authenticated WebSocket. ## Message flow 1. Send `{"text":" "}` immediately after connecting. This initialization space is not billed. 2. Send text fragments while receiving `audio` chunks. Set `flush:true` to finish a segment without closing the connection. Audio may arrive before a flush. 3. Send `{"text":""}` to end input. Keep receiving until `{"isFinal":true}`, then the connection closes. Output streams automatically. Only automatic generation is supported. There is no public flush-completion event; `isFinal` completes the whole connection. See the [TTS input streaming guide](/streaming-text) for a runnable example. parameters: voice_id: description: Voice ID from the selected model catalog, such as diana. schema: type: string bindings: ws: query: type: object properties: model_id: $ref: '#/components/schemas/inputStream_model_id' output_format: $ref: '#/components/schemas/inputStream_output_format' default: pcm_24000 language_code: type: string default: auto seed: type: integer minimum: 0 maximum: 4294967295 default: 0 inactivity_timeout: type: integer minimum: 1 maximum: 180 default: 20 auto_mode: type: boolean default: true sync_alignment: type: boolean default: false enable_ssml_parsing: type: boolean default: false apply_text_normalization: $ref: '#/components/schemas/inputStream_apply_text_normalization' default: 'off' publish: operationId: subpackage_inputStream.Stream text to speech-publish summary: Server messages message: oneOf: - $ref: >- #/components/messages/subpackage_inputStream.Stream text to speech-server-0-Audio - $ref: >- #/components/messages/subpackage_inputStream.Stream text to speech-server-1-Final - $ref: >- #/components/messages/subpackage_inputStream.Stream text to speech-server-2-Error subscribe: operationId: subpackage_inputStream.Stream text to speech-subscribe summary: Client messages message: oneOf: - $ref: >- #/components/messages/subpackage_inputStream.Stream text to speech-client-0-Initialize - $ref: >- #/components/messages/subpackage_inputStream.Stream text to speech-client-1-Text - $ref: >- #/components/messages/subpackage_inputStream.Stream text to speech-client-2-End servers: Production: url: wss://api.narilabs.com/ protocol: wss x-default: true components: messages: subpackage_inputStream.Stream text to speech-server-0-Audio: name: Audio title: Audio chunk description: >- Decode the base64 `audio` as raw PCM16 at 24 kHz. Chunks arrive in order. payload: $ref: '#/components/schemas/inputStream_Audio' subpackage_inputStream.Stream text to speech-server-1-Final: name: Final title: Connection complete description: >- The `isFinal:true` message confirms that all audio for the connection has been sent. payload: $ref: '#/components/schemas/inputStream_Final' subpackage_inputStream.Stream text to speech-server-2-Error: name: Error title: Error description: A terminal error may be followed by a non-1000 WebSocket close. payload: $ref: '#/components/schemas/inputStream_Error' subpackage_inputStream.Stream text to speech-client-0-Initialize: name: Initialize title: Initialize description: Send `{"text":" "}` once to initialize the connection. payload: $ref: '#/components/schemas/inputStream_Initialize' subpackage_inputStream.Stream text to speech-client-1-Text: name: Text title: Send text / flush description: >- Append text. Set `flush:true` to finish the current segment and keep the connection open. payload: $ref: '#/components/schemas/inputStream_Text' subpackage_inputStream.Stream text to speech-client-2-End: name: End title: End input description: Send empty text and keep receiving until `isFinal` arrives. payload: $ref: '#/components/schemas/inputStream_End' schemas: inputStream_model_id: type: string enum: - qwen3-tts - qwen3-tts-fast title: inputStream_model_id inputStream_output_format: type: string enum: - pcm_24000 default: pcm_24000 title: inputStream_output_format inputStream_apply_text_normalization: type: string enum: - 'off' default: 'off' title: inputStream_apply_text_normalization inputStream_Audio: type: object properties: audio: type: string description: >- Base64-encoded 24 kHz signed 16-bit little-endian mono PCM, without a WAV header. isFinal: type: boolean required: - audio - isFinal title: inputStream_Audio inputStream_Final: type: object properties: isFinal: type: boolean description: Entire connection completed; not emitted for each flush. required: - isFinal title: inputStream_Final inputStream_Error: type: object properties: error: type: string description: Lowercase machine-readable error code. message: type: string description: Human-readable explanation. requestId: type: string description: >- Present on Gateway-generated errors; may be omitted on worker errors. Also keep the x-request-id handshake header. required: - error - message title: inputStream_Error ChannelsInputStreamMessagesInitializeText: type: string enum: - ' ' description: >- Exactly one space. Send once immediately after connecting; it is not billed. title: ChannelsInputStreamMessagesInitializeText VoiceSettingsStability: type: string enum: - '0.5' title: VoiceSettingsStability VoiceSettingsSimilarityBoost: type: string enum: - '0.75' title: VoiceSettingsSimilarityBoost VoiceSettingsStyle: type: string enum: - '0' title: VoiceSettingsStyle VoiceSettingsSpeed: type: string enum: - '1' title: VoiceSettingsSpeed VoiceSettings: type: object properties: stability: $ref: '#/components/schemas/VoiceSettingsStability' default: 0.5 similarity_boost: $ref: '#/components/schemas/VoiceSettingsSimilarityBoost' default: 0.75 style: $ref: '#/components/schemas/VoiceSettingsStyle' default: 0 use_speaker_boost: type: boolean default: true speed: $ref: '#/components/schemas/VoiceSettingsSpeed' default: 1 title: VoiceSettings inputStream_Initialize: type: object properties: text: $ref: '#/components/schemas/ChannelsInputStreamMessagesInitializeText' description: >- Exactly one space. Send once immediately after connecting; it is not billed. voice_settings: oneOf: - $ref: '#/components/schemas/VoiceSettings' - type: 'null' description: >- Optional default settings for compatibility. Must not change after initialization. xi_api_key: type: - string - 'null' description: >- Compatibility field only; ignored for public API authentication. Use the Authorization handshake header. xi-api-key: type: - string - 'null' description: >- Compatibility field only; ignored for public API authentication. Use the Authorization handshake header. authorization: type: - string - 'null' description: >- Compatibility field only; ignored for public API authentication. Use the Authorization handshake header. required: - text title: inputStream_Initialize inputStream_Text: type: object properties: text: type: string minLength: 1 description: >- Append text verbatim. A standalone single space is a keepalive, excluded from synthesis and billing. flush: type: boolean default: false description: >- Seal pending text as a segment, drain its audio, and keep the connection open. voice_settings: oneOf: - $ref: '#/components/schemas/VoiceSettings' - type: 'null' description: >- Optional default settings for compatibility. Must not change after initialization. required: - text title: inputStream_Text ChannelsInputStreamMessagesEndText: type: string enum: - '' description: End input and drain remaining audio; keep receiving until isFinal. title: ChannelsInputStreamMessagesEndText inputStream_End: type: object properties: text: $ref: '#/components/schemas/ChannelsInputStreamMessagesEndText' description: End input and drain remaining audio; keep receiving until isFinal. voice_settings: oneOf: - $ref: '#/components/schemas/VoiceSettings' - type: 'null' description: >- Optional default settings for compatibility. Must not change after initialization. flush: type: boolean default: false description: Empty text ends input regardless of this value. required: - text title: inputStream_End ```