> ## Documentation Index > Fetch the complete documentation index at: https://docs.voxnexus.ai/llms.txt > Use this file to discover all available pages before exploring further. # STT Realtime > WebSocket Real-time Speech-to-Text Channel Through this channel, clients can: 1. Send initialization messages to configure recognition parameters 2. Send audio data streams for real-time speech recognition 3. Send command messages to control recognition flow 4. Receive recognition results (partial and final results) **Connection**: - Authentication via header: `X-Api-Key: YOUR_API_KEY` - Authentication via query: `?token=YOUR_API_KEY` **Message Format**: All messages are in JSON format, containing a `type` field to identify the message type ## AsyncAPI ````yaml api-reference/asyncapi.yaml sttRealtime id: sttRealtime title: STT Realtime description: > WebSocket Real-time Speech-to-Text Channel Through this channel, clients can: 1. Send initialization messages to configure recognition parameters 2. Send audio data streams for real-time speech recognition 3. Send command messages to control recognition flow 4. Receive recognition results (partial and final results) **Connection**: - Authentication via header: `X-Api-Key: YOUR_API_KEY` - Authentication via query: `?token=YOUR_API_KEY` **Message Format**: All messages are in JSON format, containing a `type` field to identify the message type servers: - id: production protocol: wss host: api.voxnexus.ai bindings: [] variables: [] address: /v1/stt/realtime parameters: [] bindings: [] operations: - &ref_1 id: receiveSTTClientMessages title: Receive s t t client messages type: receive messages: - &ref_3 id: STTInitMessage payload: - name: STT Initialization Message type: object properties: - name: type type: string description: Message type enumValues: - init required: true - name: model_id type: string description: Model ID (required) required: true - name: language type: string description: Language code (optional) required: false - name: format type: string description: Audio format (required) enumValues: - pcm required: true - name: sample_rate type: integer description: Sample rate (required) enumValues: - 16000 required: true - name: enable_timestamps type: boolean description: Whether to return timestamps (optional) required: false - name: enable_llm_transform type: boolean description: > Whether to enable LLM post-processing on each final transcript (optional, default false). When enabled, the server sends `llm` messages alongside `transcript` messages. **This feature requires whitelist access. Contact support@voxnexus.ai to request access.** required: false - name: llm_prompt type: string description: > LLM transform instruction (required when enable_llm_transform=true). E.g. "correct punctuation", "translate to English", "rewrite humorously". required: false - name: llm_model_id type: string description: > LLM model ID (optional, falls back to server default). Routing is prefix-based: `claude-*` models route to Anthropic; all other model IDs route to OpenAI. Examples: `claude-haiku-4-5-20251001` (Anthropic), `gpt-4.1` (OpenAI). required: false - name: llm_max_tokens type: integer description: >- Maximum output tokens for LLM transform (optional, falls back to server default) required: false - name: llm_mode type: string description: > LLM processing mode (optional, default: per_segment). - `per_segment`: LLM runs on each final sentence as it arrives (low latency, real-time). - `post_flush`: LLM runs once on the full accumulated text after flush (full context). enumValues: - per_segment - post_flush required: false - name: llm_post_flush type: boolean description: > When llm_mode=per_segment, also run a full-text LLM pass after flush (optional, default false). Produces both per-sentence `llm` messages (with segment_id) and a final full-text `llm` message (without segment_id) after flush completes. required: false - name: llm_post_flush_prompt type: string description: | Separate prompt for the post-flush full-text pass (optional). Falls back to llm_prompt when not specified. required: false headers: [] jsonPayloadSchema: type: object description: STT WebSocket initialization message (sent by client) properties: type: type: string enum: - init description: Message type x-parser-schema-id: model_id: type: string description: Model ID (required) example: vn-stt-basic x-parser-schema-id: language: type: string description: Language code (optional) example: en-US x-parser-schema-id: format: type: string description: Audio format (required) enum: - pcm example: pcm x-parser-schema-id: sample_rate: type: integer description: Sample rate (required) enum: - 16000 default: 16000 example: 16000 x-parser-schema-id: enable_timestamps: type: boolean description: Whether to return timestamps (optional) default: false x-parser-schema-id: enable_llm_transform: type: boolean description: > Whether to enable LLM post-processing on each final transcript (optional, default false). When enabled, the server sends `llm` messages alongside `transcript` messages. **This feature requires whitelist access. Contact support@voxnexus.ai to request access.** default: false x-parser-schema-id: llm_prompt: type: string description: > LLM transform instruction (required when enable_llm_transform=true). E.g. "correct punctuation", "translate to English", "rewrite humorously". example: Correct punctuation and remove filler words x-parser-schema-id: llm_model_id: type: string description: > LLM model ID (optional, falls back to server default). Routing is prefix-based: `claude-*` models route to Anthropic; all other model IDs route to OpenAI. Examples: `claude-haiku-4-5-20251001` (Anthropic), `gpt-4.1` (OpenAI). example: claude-haiku-4-5-20251001 x-parser-schema-id: llm_max_tokens: type: integer description: >- Maximum output tokens for LLM transform (optional, falls back to server default) minimum: 1 maximum: 4096 example: 1024 x-parser-schema-id: llm_mode: type: string description: > LLM processing mode (optional, default: per_segment). - `per_segment`: LLM runs on each final sentence as it arrives (low latency, real-time). - `post_flush`: LLM runs once on the full accumulated text after flush (full context). enum: - per_segment - post_flush default: per_segment x-parser-schema-id: llm_post_flush: type: boolean description: > When llm_mode=per_segment, also run a full-text LLM pass after flush (optional, default false). Produces both per-sentence `llm` messages (with segment_id) and a final full-text `llm` message (without segment_id) after flush completes. default: false x-parser-schema-id: llm_post_flush_prompt: type: string description: | Separate prompt for the post-flush full-text pass (optional). Falls back to llm_prompt when not specified. example: Summarize the key points in three bullet points x-parser-schema-id: required: - type - model_id - format - sample_rate x-parser-schema-id: title: STT Initialization Message example: |- { "type": "", "model_id": "", "language": "", "format": "", "sample_rate": 123, "enable_timestamps": true, "enable_llm_transform": true, "llm_prompt": "", "llm_model_id": "", "llm_max_tokens": 123, "llm_mode": "", "llm_post_flush": true, "llm_post_flush_prompt": "" } bindings: [] extensions: - id: x-parser-unique-object-id value: STTInitMessage - id: x-parser-message-name value: STTInitMessage - &ref_4 id: STTAudioMessage payload: - name: STT Audio Message type: object properties: - name: type type: string description: Message type enumValues: - audio required: true - name: data type: string description: Base64-encoded audio data required: true headers: [] jsonPayloadSchema: type: object description: STT WebSocket audio data message (sent by client) properties: type: type: string enum: - audio description: Message type x-parser-schema-id: data: type: string format: byte description: Base64-encoded audio data x-parser-schema-id: required: - type - data x-parser-schema-id: title: STT Audio Message example: |- { "type": "", "data": "" } bindings: [] extensions: - id: x-parser-unique-object-id value: STTAudioMessage - id: x-parser-message-name value: STTAudioMessage - &ref_5 id: STTCommandMessage payload: - name: STT Command Message type: object properties: - name: type type: string description: Message type enumValues: - command required: true - name: command type: string description: >- Command type: 'flush' tells the server that no more audio will be sent. The server will respond with a 'flush_done' message after recognition is complete. Subsequent audio will start a new recognition session. enumValues: - flush required: true headers: [] jsonPayloadSchema: type: object description: STT WebSocket command message (sent by client) properties: type: type: string enum: - command description: Message type x-parser-schema-id: command: type: string enum: - flush description: >- Command type: 'flush' tells the server that no more audio will be sent. The server will respond with a 'flush_done' message after recognition is complete. Subsequent audio will start a new recognition session. x-parser-schema-id: required: - type - command x-parser-schema-id: title: STT Command Message example: |- { "type": "", "command": "" } bindings: [] extensions: - id: x-parser-unique-object-id value: STTCommandMessage - id: x-parser-message-name value: STTCommandMessage bindings: [] extensions: &ref_0 - id: x-parser-unique-object-id value: sttRealtime - &ref_2 id: sendSTTServerMessages title: Send s t t server messages type: send messages: - &ref_6 id: STTReadyMessage payload: - name: STT Ready Message type: object properties: - name: type type: string description: Message type enumValues: - ready required: true - name: request_id type: string description: Request ID required: true - name: language type: string description: Language code required: false - name: format type: string description: Audio format required: true - name: sample_rate type: integer description: Sample rate required: true - name: enable_timestamps type: boolean description: Whether timestamps are enabled required: false headers: [] jsonPayloadSchema: type: object description: STT WebSocket ready message (sent by server) properties: type: type: string enum: - ready description: Message type x-parser-schema-id: request_id: type: string description: Request ID example: req_1234567890 x-parser-schema-id: language: type: string description: Language code example: en-US x-parser-schema-id: format: type: string description: Audio format example: pcm x-parser-schema-id: sample_rate: type: integer description: Sample rate example: 16000 x-parser-schema-id: enable_timestamps: type: boolean description: Whether timestamps are enabled example: false x-parser-schema-id: required: - type - request_id - format - sample_rate x-parser-schema-id: title: STT Ready Message example: |- { "type": "", "request_id": "", "language": "", "format": "", "sample_rate": 123, "enable_timestamps": true } bindings: [] extensions: - id: x-parser-unique-object-id value: STTReadyMessage - id: x-parser-message-name value: STTReadyMessage - &ref_7 id: STTTranscriptMessage payload: - name: STT Transcript Message type: object properties: - name: type type: string description: Message type enumValues: - transcript required: true - name: request_id type: string description: Request ID required: true - name: segment_id type: string description: > Unique sentence-level ID (only present when is_final=true and LLM transform is enabled). Used to correlate subsequent `llm` messages to this specific final transcript segment. required: false - name: text type: string description: Recognized text required: true - name: is_final type: boolean description: >- Whether this is a complete sentence (false=partial result, true=sentence complete) required: true - name: language type: string description: >- Detected language code (if language detection is enabled), E.g. en, en-US required: false - name: confidence type: number description: Confidence score 0.0-1.0 (only valid when is_final=true) required: false - name: offset type: integer description: Text offset in audio stream (milliseconds) required: false - name: duration type: integer description: >- Duration of corresponding audio (milliseconds, only valid when is_final=true) required: false - name: words type: array description: >- Word-level information array (only valid when is_final=true and timestamps are enabled) required: false properties: - name: word type: string description: Word required: true - name: offset type: integer description: Word offset in audio stream (milliseconds) required: false - name: duration type: integer description: Word duration (milliseconds) required: false - name: confidence type: number description: Confidence score 0.0-1.0 required: false headers: [] jsonPayloadSchema: type: object description: >- STT WebSocket transcript result message (sent by server). The is_final field distinguishes between partial results (false) and complete sentences (true). properties: type: type: string enum: - transcript description: Message type x-parser-schema-id: request_id: type: string description: Request ID example: req_1234567890 x-parser-schema-id: segment_id: type: string format: uuid description: > Unique sentence-level ID (only present when is_final=true and LLM transform is enabled). Used to correlate subsequent `llm` messages to this specific final transcript segment. example: 550e8400-e29b-41d4-a716-446655440000 x-parser-schema-id: text: type: string description: Recognized text example: Hello, this is a test message. x-parser-schema-id: is_final: type: boolean description: >- Whether this is a complete sentence (false=partial result, true=sentence complete) example: true x-parser-schema-id: language: type: string description: >- Detected language code (if language detection is enabled), E.g. en, en-US example: en x-parser-schema-id: confidence: type: number format: float description: Confidence score 0.0-1.0 (only valid when is_final=true) minimum: 0 maximum: 1 example: 0.95 x-parser-schema-id: offset: type: integer description: Text offset in audio stream (milliseconds) example: 0 x-parser-schema-id: duration: type: integer description: >- Duration of corresponding audio (milliseconds, only valid when is_final=true) example: 2500 x-parser-schema-id: words: type: array description: >- Word-level information array (only valid when is_final=true and timestamps are enabled) items: title: STT Word Information type: object properties: word: type: string description: Word example: hello x-parser-schema-id: offset: type: integer description: Word offset in audio stream (milliseconds) example: 0 x-parser-schema-id: duration: type: integer description: Word duration (milliseconds) example: 500 x-parser-schema-id: confidence: type: number format: float description: Confidence score 0.0-1.0 minimum: 0 maximum: 1 example: 0.98 x-parser-schema-id: required: - word x-parser-schema-id: STTWordInfo x-parser-schema-id: required: - type - request_id - text - is_final x-parser-schema-id: title: STT Transcript Message example: |- { "type": "", "request_id": "", "segment_id": "", "text": "", "is_final": true, "language": "", "confidence": 123, "offset": 123, "duration": 123, "words": { "word": "", "offset": 123, "duration": 123, "confidence": 123 } } bindings: [] extensions: - id: x-parser-unique-object-id value: STTTranscriptMessage - id: x-parser-message-name value: STTTranscriptMessage - &ref_8 id: STTLLMMessage payload: - name: STT LLM Transform Message type: object properties: - name: type type: string description: Message type enumValues: - llm required: true - name: request_id type: string description: Request ID required: true - name: segment_id type: string description: > Links this LLM output to the corresponding `transcript` message (per_segment mode only). Absent for full-text post_flush pass. required: false - name: delta type: string description: Incremental LLM output text chunk (streaming token) required: true - name: is_final type: boolean description: > Whether this is the last chunk for this segment's LLM output. Clients should concatenate all delta values until is_final=true. required: true headers: [] jsonPayloadSchema: type: object description: > LLM transform streaming output message (sent by server when enable_llm_transform=true). Each final transcript segment triggers a sequence of `llm` messages streaming the LLM output incrementally (delta chunks), terminated by a message with is_final=true. - With `segment_id`: corresponds to a specific per-segment LLM pass (llm_mode=per_segment). - Without `segment_id`: corresponds to a full-text LLM pass (llm_mode=post_flush, or per_segment+llm_post_flush after all segment passes complete). properties: type: type: string enum: - llm description: Message type x-parser-schema-id: request_id: type: string description: Request ID example: req_1234567890 x-parser-schema-id: segment_id: type: string format: uuid description: > Links this LLM output to the corresponding `transcript` message (per_segment mode only). Absent for full-text post_flush pass. example: 550e8400-e29b-41d4-a716-446655440000 x-parser-schema-id: delta: type: string description: Incremental LLM output text chunk (streaming token) example: Hello, x-parser-schema-id: is_final: type: boolean description: | Whether this is the last chunk for this segment's LLM output. Clients should concatenate all delta values until is_final=true. example: false x-parser-schema-id: required: - type - request_id - delta - is_final x-parser-schema-id: title: STT LLM Transform Message example: |- { "type": "", "request_id": "", "segment_id": "", "delta": "", "is_final": true } bindings: [] extensions: - id: x-parser-unique-object-id value: STTLLMMessage - id: x-parser-message-name value: STTLLMMessage - &ref_9 id: STTFlushDoneMessage payload: - name: STT Flush Done Message type: object properties: - name: type type: string description: Message type enumValues: - flush_done required: true - name: request_id type: string description: Request ID required: true headers: [] jsonPayloadSchema: type: object description: >- STT WebSocket flush completion message (sent by server after receiving 'flush' command and completing recognition) properties: type: type: string enum: - flush_done description: Message type x-parser-schema-id: request_id: type: string description: Request ID example: req_1234567890 x-parser-schema-id: required: - type - request_id x-parser-schema-id: title: STT Flush Done Message example: |- { "type": "", "request_id": "" } bindings: [] extensions: - id: x-parser-unique-object-id value: STTFlushDoneMessage - id: x-parser-message-name value: STTFlushDoneMessage - &ref_10 id: STTErrorMessage payload: - name: STT Error Message type: object properties: - name: type type: string description: Message type enumValues: - error required: true - name: error type: string description: Error description required: true - name: code type: string description: Error code required: true - name: request_id type: string description: Request ID required: false headers: [] jsonPayloadSchema: type: object description: STT WebSocket error message (sent by server) properties: type: type: string enum: - error description: Message type x-parser-schema-id: error: type: string description: Error description example: Invalid audio format x-parser-schema-id: code: type: string description: Error code example: UNSUPPORTED_FORMAT x-parser-schema-id: request_id: type: string description: Request ID example: req_1234567890 x-parser-schema-id: required: - type - error - code x-parser-schema-id: title: STT Error Message example: |- { "type": "", "error": "", "code": "", "request_id": "" } bindings: [] extensions: - id: x-parser-unique-object-id value: STTErrorMessage - id: x-parser-message-name value: STTErrorMessage bindings: [] extensions: *ref_0 sendOperations: - *ref_1 receiveOperations: - *ref_2 sendMessages: - *ref_3 - *ref_4 - *ref_5 receiveMessages: - *ref_6 - *ref_7 - *ref_8 - *ref_9 - *ref_10 extensions: - id: x-parser-unique-object-id value: sttRealtime securitySchemes: - id: apiKeyAuth name: token type: httpApiKey description: API Key passed as query parameter, e.g. ?token=YOUR_API_KEY in: query extensions: [] ````