> ## Documentation Index > Fetch the complete documentation index at: https://voice.ai/llms.txt > Use this file to discover all available pages before exploring further. # Multi-Context WebSocket > Multiple concurrent TTS streams over a single WebSocket connection. Each context has its own voice and settings. For multiple generations, use this instead of `/stream`. **Authentication:** `Authorization: Bearer YOUR_API_KEY` ## AsyncAPI ````yaml docs/asyncapi.json /api/v1/tts/multi-stream id: /api/v1/tts/multi-stream title: Multi-Context WebSocket description: >- Multiple concurrent TTS streams over a single WebSocket connection. Each context has its own voice and settings. For multiple generations, use this instead of `/stream`. **Authentication:** `Authorization: Bearer YOUR_API_KEY` servers: - id: production protocol: wss host: dev.voice.ai bindings: [] variables: [] address: /api/v1/tts/multi-stream parameters: [] bindings: [] operations: - &ref_1 id: sendInitMulti title: Send Context Initialization description: >- Send the first message to a context_id to initialize that context with voice, model, and language settings type: receive messages: - &ref_6 id: initMessage contentType: application/json payload: - name: Context Initialization Message description: First message to a context_id to set up that context type: object properties: - name: context_id type: string description: Context identifier (auto-generated if omitted) required: false - name: voice_id type: string description: Voice ID to use. Omit to use the default built-in voice. required: false - name: text type: string description: Text to synthesize (buffered until flush) required: true - name: language type: string description: >- Language code (ISO 639-1 format). Supported: en, ca, sv, es, fr, de, it, pt, pl, ru, nl. enumValues: - en - ca - sv - es - fr - de - it - pt - pl - ru - nl required: false - name: model type: string description: >- TTS model to use. If not provided, automatically selected based on language. English uses voiceai-tts-v1-latest; other languages use voiceai-tts-multilingual-v1-latest. Lite models are English-only and must be selected explicitly. enumValues: - voiceai-tts-v1-latest - voiceai-tts-v1-2026-02-10 - voiceai-tts-lite-v1-latest - voiceai-tts-lite-v1-2026-04-15 - voiceai-tts-multilingual-v1-latest - voiceai-tts-multilingual-v1-2026-02-10 required: false - name: audio_format type: string description: >- Audio output format. Basic formats (mp3, wav, pcm) use 32kHz sample rate. Format-specific options allow control over sample rate and bitrate. All PCM formats (pcm, pcm_*) use 16-bit signed little-endian mono. enumValues: - mp3 - wav - pcm - alaw_8000 - mp3_22050_32 - mp3_24000_48 - mp3_44100_32 - mp3_44100_64 - mp3_44100_96 - mp3_44100_128 - mp3_44100_192 - opus_48000_32 - opus_48000_64 - opus_48000_96 - opus_48000_128 - opus_48000_192 - pcm_8000 - pcm_16000 - pcm_22050 - pcm_24000 - pcm_32000 - pcm_44100 - pcm_48000 - ulaw_8000 - wav_16000 - wav_22050 - wav_24000 required: false - name: temperature type: number description: Sampling temperature (0.0-2.0) required: false - name: top_p type: number description: Nucleus sampling (0.0-1.0) required: false - name: dictionary_id type: string description: Optional managed pronunciation dictionary identifier. required: false - name: dictionary_version type: integer description: >- Optional managed dictionary version. Defaults to latest when omitted. required: false - name: delivery_mode type: string description: >- Delivery cadence mode: 'raw' (default) or 'paced'. Paced delivery is applied only to PCM-based outputs ('pcm', 'pcm_*', 'ulaw_8000', 'alaw_8000'); other formats fall back to raw cadence. enumValues: - raw - paced required: false - name: flush type: boolean description: Trigger audio generation from buffer required: false - name: auto_close type: boolean description: >- If true, automatically close this context after flush completes. Useful for fire-and-forget TTS generation where you want to release the concurrent generation slot immediately. Server sends context_closed after is_last when this is set. required: false headers: [] jsonPayloadSchema: type: object properties: context_id: type: string description: Context identifier (auto-generated if omitted) x-parser-schema-id: voice_id: type: string description: Voice ID to use. Omit to use the default built-in voice. x-parser-schema-id: text: type: string description: Text to synthesize (buffered until flush) x-parser-schema-id: language: type: string description: >- Language code (ISO 639-1 format). Supported: en, ca, sv, es, fr, de, it, pt, pl, ru, nl. default: en enum: - en - ca - sv - es - fr - de - it - pt - pl - ru - nl x-parser-schema-id: model: type: string description: >- TTS model to use. If not provided, automatically selected based on language. English uses voiceai-tts-v1-latest; other languages use voiceai-tts-multilingual-v1-latest. Lite models are English-only and must be selected explicitly. enum: - voiceai-tts-v1-latest - voiceai-tts-v1-2026-02-10 - voiceai-tts-lite-v1-latest - voiceai-tts-lite-v1-2026-04-15 - voiceai-tts-multilingual-v1-latest - voiceai-tts-multilingual-v1-2026-02-10 x-parser-schema-id: audio_format: type: string description: >- Audio output format. Basic formats (mp3, wav, pcm) use 32kHz sample rate. Format-specific options allow control over sample rate and bitrate. All PCM formats (pcm, pcm_*) use 16-bit signed little-endian mono. enum: - mp3 - wav - pcm - alaw_8000 - mp3_22050_32 - mp3_24000_48 - mp3_44100_32 - mp3_44100_64 - mp3_44100_96 - mp3_44100_128 - mp3_44100_192 - opus_48000_32 - opus_48000_64 - opus_48000_96 - opus_48000_128 - opus_48000_192 - pcm_8000 - pcm_16000 - pcm_22050 - pcm_24000 - pcm_32000 - pcm_44100 - pcm_48000 - ulaw_8000 - wav_16000 - wav_22050 - wav_24000 default: mp3 x-enum-descriptions: mp3: MP3 format (32kHz) wav: WAV format (32kHz) pcm: PCM format (32kHz) alaw_8000: A-law telephony format (8kHz) mp3_22050_32: MP3 at 22.05kHz, 32kbps mp3_24000_48: MP3 at 24kHz, 48kbps mp3_44100_32: MP3 at 44.1kHz, 32kbps mp3_44100_64: MP3 at 44.1kHz, 64kbps mp3_44100_96: MP3 at 44.1kHz, 96kbps mp3_44100_128: MP3 at 44.1kHz, 128kbps mp3_44100_192: MP3 at 44.1kHz, 192kbps opus_48000_32: Opus at 48kHz, 32kbps opus_48000_64: Opus at 48kHz, 64kbps opus_48000_96: Opus at 48kHz, 96kbps opus_48000_128: Opus at 48kHz, 128kbps opus_48000_192: Opus at 48kHz, 192kbps pcm_8000: PCM at 8kHz pcm_16000: PCM at 16kHz pcm_22050: PCM at 22.05kHz pcm_24000: PCM at 24kHz pcm_32000: PCM at 32kHz pcm_44100: PCM at 44.1kHz pcm_48000: PCM at 48kHz ulaw_8000: μ-law telephony format (8kHz) wav_16000: WAV at 16kHz wav_22050: WAV at 22.05kHz wav_24000: WAV at 24kHz x-parser-schema-id: temperature: type: number description: Sampling temperature (0.0-2.0) default: 1 minimum: 0 maximum: 2 x-parser-schema-id: top_p: type: number description: Nucleus sampling (0.0-1.0) default: 0.8 minimum: 0 maximum: 1 x-parser-schema-id: dictionary_id: type: string description: Optional managed pronunciation dictionary identifier. x-parser-schema-id: dictionary_version: type: integer description: >- Optional managed dictionary version. Defaults to latest when omitted. minimum: 1 x-parser-schema-id: delivery_mode: type: string description: >- Delivery cadence mode: 'raw' (default) or 'paced'. Paced delivery is applied only to PCM-based outputs ('pcm', 'pcm_*', 'ulaw_8000', 'alaw_8000'); other formats fall back to raw cadence. enum: - raw - paced default: raw x-parser-schema-id: flush: type: boolean description: Trigger audio generation from buffer default: false x-parser-schema-id: auto_close: type: boolean description: >- If true, automatically close this context after flush completes. Useful for fire-and-forget TTS generation where you want to release the concurrent generation slot immediately. Server sends context_closed after is_last when this is set. default: false x-parser-schema-id: required: - text x-parser-schema-id: title: Context Initialization Message description: First message to a context_id to set up that context example: |- { "context_id": "", "voice_id": "", "text": "", "language": "", "model": "", "audio_format": "", "temperature": 123, "top_p": 123, "dictionary_id": "", "dictionary_version": 123, "delivery_mode": "", "flush": true, "auto_close": true } bindings: [] extensions: - id: x-parser-unique-object-id value: initMessage bindings: [] extensions: &ref_0 - id: x-parser-unique-object-id value: /api/v1/tts/multi-stream - &ref_2 id: sendTextMulti title: Send Text to Context description: >- Send text-only messages to an existing context (voice settings are remembered for that context) type: receive messages: - &ref_7 id: textMessage contentType: application/json payload: - name: Text-Only Message description: >- Subsequent messages to an existing context (text-only, no params). Can include close_context or close_socket to control context/connection lifecycle. type: object properties: - name: context_id type: string description: >- Context identifier. Required to identify which context this message belongs to. required: true - name: text type: string description: >- Text to generate speech for. Will be accumulated until flush. Optional if close_context is True. required: false - name: flush type: boolean description: >- If true, generate speech for all buffered text. Optional if close_context is True. required: false - name: close_context type: boolean description: >- If true, close the specified context after processing this message. Can be combined with text/flush - flush will complete first, then context closes. required: false - name: close_socket type: boolean description: >- If true, close the entire WebSocket connection after processing this message. required: false - name: auto_close type: boolean description: >- If true, automatically close this context after flush completes. Useful for fire-and-forget TTS generation where you want to release the concurrent generation slot immediately. Server sends context_closed after is_last when this is set. required: false headers: [] jsonPayloadSchema: type: object properties: context_id: type: string description: >- Context identifier. Required to identify which context this message belongs to. x-parser-schema-id: text: type: string description: >- Text to generate speech for. Will be accumulated until flush. Optional if close_context is True. default: '' x-parser-schema-id: flush: type: boolean description: >- If true, generate speech for all buffered text. Optional if close_context is True. default: false x-parser-schema-id: close_context: type: boolean description: >- If true, close the specified context after processing this message. Can be combined with text/flush - flush will complete first, then context closes. default: false x-parser-schema-id: close_socket: type: boolean description: >- If true, close the entire WebSocket connection after processing this message. default: false x-parser-schema-id: auto_close: type: boolean description: >- If true, automatically close this context after flush completes. Useful for fire-and-forget TTS generation where you want to release the concurrent generation slot immediately. Server sends context_closed after is_last when this is set. default: false x-parser-schema-id: required: - context_id x-parser-schema-id: title: Text-Only Message description: >- Subsequent messages to an existing context (text-only, no params). Can include close_context or close_socket to control context/connection lifecycle. example: |- { "context_id": "", "text": "", "flush": true, "close_context": true, "close_socket": true, "auto_close": true } bindings: [] extensions: - id: x-parser-unique-object-id value: textMessage bindings: [] extensions: *ref_0 - &ref_5 id: receiveAudioMulti title: Receive Audio Chunks description: >- Receive base64-encoded audio chunks with context_id as they are generated (interleaved for multiple contexts). After each flush completes, you will receive an is_last message. The same context can be reused for multiple flushes. type: send messages: - &ref_10 id: audioChunk contentType: application/json payload: - name: Audio Chunk Response description: Base64-encoded audio chunk with context_id (streamed immediately) type: object properties: - name: audio type: string description: Base64-encoded audio chunk (32kHz sample rate) required: true - name: context_id type: string description: Context identifier this audio belongs to required: true headers: [] jsonPayloadSchema: type: object properties: audio: type: string description: Base64-encoded audio chunk (32kHz sample rate) format: byte x-parser-schema-id: context_id: type: string description: Context identifier this audio belongs to x-parser-schema-id: required: - audio - context_id x-parser-schema-id: title: Audio Chunk Response description: Base64-encoded audio chunk with context_id (streamed immediately) example: |- { "audio": "", "context_id": "" } bindings: [] extensions: - id: x-parser-unique-object-id value: audioChunk - &ref_11 id: completionMessage contentType: application/json payload: - name: Flush Completion Signal description: >- Sent after all audio chunks for a flush operation (separate message). Indicates inference is complete for this flush. The context remains active and can receive more text/flushes. Each flush generates its own is_last message. type: object properties: - name: is_last type: boolean description: >- Indicates all audio chunks for this flush have been sent. Sent after each flush completes. The context remains active for more flushes. required: true - name: context_id type: string description: Context identifier this completion belongs to required: true headers: [] jsonPayloadSchema: type: object properties: is_last: type: boolean description: >- Indicates all audio chunks for this flush have been sent. Sent after each flush completes. The context remains active for more flushes. const: true x-parser-schema-id: context_id: type: string description: Context identifier this completion belongs to x-parser-schema-id: required: - is_last - context_id x-parser-schema-id: title: Flush Completion Signal description: >- Sent after all audio chunks for a flush operation (separate message). Indicates inference is complete for this flush. The context remains active and can receive more text/flushes. Each flush generates its own is_last message. example: |- { "is_last": true, "context_id": "" } bindings: [] extensions: - id: x-parser-unique-object-id value: completionMessage - &ref_12 id: contextClosedMessage contentType: application/json payload: - name: Context Closure Confirmation description: >- Sent when a context is explicitly closed via close_context. This is a separate message from is_last to distinguish context closure from flush completion. type: object properties: - name: context_closed type: boolean description: >- Indicates the context has been closed. Sent as confirmation after receiving a close_context message. required: true - name: context_id type: string description: Context identifier that was closed required: true headers: [] jsonPayloadSchema: type: object properties: context_closed: type: boolean description: >- Indicates the context has been closed. Sent as confirmation after receiving a close_context message. const: true x-parser-schema-id: context_id: type: string description: Context identifier that was closed x-parser-schema-id: required: - context_closed - context_id x-parser-schema-id: title: Context Closure Confirmation description: >- Sent when a context is explicitly closed via close_context. This is a separate message from is_last to distinguish context closure from flush completion. example: |- { "context_closed": true, "context_id": "" } bindings: [] extensions: - id: x-parser-unique-object-id value: contextClosedMessage - &ref_13 id: errorMessage contentType: application/json payload: - name: Error Response description: Error message from server with context_id type: object properties: - name: error type: string description: Error message required: true - name: context_id type: string description: Context identifier (if applicable) required: false headers: [] jsonPayloadSchema: type: object properties: error: type: string description: Error message x-parser-schema-id: context_id: type: string description: Context identifier (if applicable) nullable: true x-parser-schema-id: required: - error x-parser-schema-id: title: Error Response description: Error message from server with context_id example: |- { "error": "", "context_id": "" } bindings: [] extensions: - id: x-parser-unique-object-id value: errorMessage bindings: [] extensions: *ref_0 - &ref_3 id: closeContextMulti title: Close Context description: >- Close a specific context. Server responds with context_closed message to confirm the context is closed. type: receive messages: - &ref_8 id: closeContext contentType: application/json payload: - name: Close Context Message description: >- Close a specific context. Can be sent as a standalone message or included in any message. Server responds with context_closed message to confirm the context is closed. type: object properties: - name: context_id type: string description: Context identifier to close required: true - name: close_context type: boolean description: >- Must be true to close this context. Can be included in any message type. Server will send context_closed message as confirmation. required: true headers: [] jsonPayloadSchema: type: object properties: context_id: type: string description: Context identifier to close x-parser-schema-id: close_context: type: boolean description: >- Must be true to close this context. Can be included in any message type. Server will send context_closed message as confirmation. const: true x-parser-schema-id: required: - context_id - close_context x-parser-schema-id: title: Close Context Message description: >- Close a specific context. Can be sent as a standalone message or included in any message. Server responds with context_closed message to confirm the context is closed. example: |- { "context_id": "", "close_context": true } bindings: [] extensions: - id: x-parser-unique-object-id value: closeContext bindings: [] extensions: *ref_0 - &ref_4 id: closeSocketMulti title: Close WebSocket Connection description: >- Close the entire WebSocket connection and all contexts. Server closes the connection after processing. type: receive messages: - &ref_9 id: closeSocket contentType: application/json payload: - name: Close Socket Message description: >- Close entire WebSocket connection and all contexts. Can be sent as a standalone message or included in any message. Server closes the connection after processing. type: object properties: - name: close_socket type: boolean description: >- Must be true to close entire connection. Can be included in any message type. All contexts are closed and the WebSocket connection is terminated. required: true headers: [] jsonPayloadSchema: type: object properties: close_socket: type: boolean description: >- Must be true to close entire connection. Can be included in any message type. All contexts are closed and the WebSocket connection is terminated. const: true x-parser-schema-id: required: - close_socket x-parser-schema-id: title: Close Socket Message description: >- Close entire WebSocket connection and all contexts. Can be sent as a standalone message or included in any message. Server closes the connection after processing. example: |- { "close_socket": true } bindings: [] extensions: - id: x-parser-unique-object-id value: closeSocket bindings: [] extensions: *ref_0 sendOperations: - *ref_1 - *ref_2 - *ref_3 - *ref_4 receiveOperations: - *ref_5 sendMessages: - *ref_6 - *ref_7 - *ref_8 - *ref_9 receiveMessages: - *ref_10 - *ref_11 - *ref_12 - *ref_13 extensions: - id: x-parser-unique-object-id value: /api/v1/tts/multi-stream securitySchemes: - id: bearerAuth name: bearerAuth type: http description: >- API key (vk_*). Include in Authorization header: 'Bearer YOUR_API_KEY'. Required for all WebSocket connections. See the [Authentication guide](/docs/guides/authentication) for details. scheme: bearer extensions: [] ````