> ## Documentation Index > Fetch the complete documentation index at: https://docs.inworld.ai/llms.txt > Use this file to discover all available pages before exploring further. # Transcribe audio (WebSocket) > Bidirectional streaming API for real-time speech-to-text transcription over WebSocket. This method listens for streaming audio input and returns recognized text chunks one by one as soon as they are ready. Audio chunks are expected to be a part of a single voice input. Suitable for streaming live conversations, microphone input, or other streaming audio sources. To use the API: - Send a `transcribeConfig` message first to configure the session (model, language, audio encoding, etc.). - Stream `audioChunk` messages containing raw audio bytes. - Receive `transcription` results as they become available, including both interim (partial) and final results. - Listen for `speechStarted` and `speechStopped` events to detect voice activity changes. - Optionally send `endTurn` to signal end of a speaker's turn. - Send `closeStream` when done. ## AsyncAPI ````yaml api-reference/sttAPI/transcribe-stream-websocket.json sttStream id: sttStream title: Stt stream description: Primary WebSocket channel for bidirectional speech-to-text streaming. servers: - id: production protocol: wss host: api.inworld.ai bindings: [] variables: [] address: /stt/v1/transcribe:streamBidirectional parameters: [] bindings: [] operations: - &ref_1 id: sendRequest title: Send request type: receive messages: - &ref_3 id: transcribeConfig payload: - name: transcribeConfig description: >- Configure the transcription session. Must be the first message sent. Contains model selection, audio format settings, and optional feature configurations. type: object properties: - name: modelId type: string description: >- The identifier of the model to use for transcription. Format: "{provider}/{model-name}". Available models: - `inworld/inworld-stt-1` — Inworld first-party - `assemblyai/universal-streaming-multilingual` — AssemblyAI multilingual - `assemblyai/universal-streaming-english` — AssemblyAI English - `assemblyai/u3-rt-pro` — AssemblyAI high-accuracy - `assemblyai/whisper-rt` — AssemblyAI Whisper real-time - `soniox/stt-rt-v4` — Soniox real-time - `soniox/stt-rt-v5` — Soniox real-time (latest generation) - `deepgram/flux-general-en` — Deepgram Flux English conversational - `deepgram/flux-general-multi` — Deepgram Flux multilingual conversational See [STT Introduction](/stt/overview) for the full model catalogue. required: true - name: audioEncoding type: string description: |- Supported audio encoding formats. - `AUDIO_ENCODING_UNSPECIFIED`: Not specified. Will return an error. - `AUTO_DETECT`: Automatically detect audio encoding from the audio header. - `LINEAR16`: Uncompressed 16-bit signed little-endian samples (Linear PCM). - `MP3`: MP3 audio. Not supported for streaming transcription. - `OGG_OPUS`: Opus encoded audio wrapped in an OGG container. Not supported for streaming transcription. - `FLAC`: FLAC encoded audio. Lossless format. Not supported for streaming transcription. enumValues: - AUDIO_ENCODING_UNSPECIFIED - AUTO_DETECT - LINEAR16 - MP3 - OGG_OPUS - FLAC required: true - name: language type: string description: >- Language hint in ISO 639 format (e.g., "en", "ja"). Biases the model toward the specified language during automatic language detection. BCP-47 codes (e.g., "en-US") are also accepted and converted to the base language code. For the Inworld first-party model (`inworld/inworld-stt-1`), the hint additionally constrains the output script for English, Chinese, Cantonese, Japanese, Korean, Russian, and Hindi (e.g. selecting `en` keeps output in Latin script). See [Language Support](/stt/languages) for the full list of supported languages. required: false - name: sampleRateHertz type: integer description: >- Sample rate of the audio data in Hertz. Required when the sample rate cannot be inferred from the audio header (e.g., raw PCM streams). Default: 16000. required: false - name: numberOfChannels type: integer description: >- Number of channels in the audio data. Required when the number of channels cannot be inferred from the audio header (e.g., raw PCM streams). Default: 1. required: false - name: inactivityTimeoutSeconds type: integer description: >- Inactivity timeout in seconds. If the client is silent for this duration, the transcription will be stopped. required: false - name: endOfTurnConfidenceThreshold type: number description: >- Confidence threshold for end-of-turn prediction. Higher values reduce false-positives. Range: [0.0, 1.0]. Default: 0.5. See the [Turn Detection guide](/stt/turn-detection) for tuning guidance. required: false - name: prompts type: array description: >- Custom vocabulary / key terms. An array of context strings (names, jargon, acronyms) that bias the model toward recognizing these terms. This is a soft bias that helps with ambiguous or uncommon words; it is not a hard keyword lock and does not force exact output. Supported across models — the unified field maps to each provider's mechanism (Groq `prompt`, AssemblyAI `keyterms_prompt`, Soniox `context`). Use letters, digits, spaces, and basic punctuation; other characters (such as #, /, @, or |) are rejected by the gateway with INVALID_ARGUMENT (code 3). required: false properties: - name: item type: string required: false - name: includeWordTimestamps type: boolean description: >- If true, includes per-word timing information in the response. Available for AssemblyAI and Soniox models. Not yet supported for `inworld/inworld-stt-1`. required: false - name: groqConfig type: object description: Configuration for Groq streaming STT models. required: false properties: - name: temperature type: number description: >- Temperature for the model. Controls randomness in predictions. Higher values produce more varied output. Range: [0.0, 1.0]. required: false - name: assemblyaiConfig type: object description: Configuration for AssemblyAI streaming STT models. required: false properties: - name: minEndOfTurnSilenceWhenConfident type: integer description: >- Minimum silence duration when confidence is high (milliseconds). required: false - name: maxTurnSilence type: integer description: >- Maximum allowed silence before forcing a turn boundary (milliseconds). required: false - name: vadThreshold type: number description: >- Voice activity detection threshold. Range: [0.0, 1.0]. Default: 0.5. required: false - name: prompt type: string description: >- Custom transcription instructions for the model. Works only for Universal-3 Pro Streaming. required: false - name: inworldSttV1Config type: object description: >- Configuration for Inworld STT 1 models. Set `vadThreshold` to 0 to disable server-side turn detection and control turn boundaries manually via `endTurn` — see the [Turn Detection guide](/stt/turn-detection). required: false properties: - name: minEndOfTurnSilenceWhenConfident type: integer description: >- Minimum silence duration when confidence is high (milliseconds). required: false - name: vadThreshold type: number description: >- Voice activity detection threshold. Range: [0.0, 1.0]. Default: 0.5. required: false - name: sonioxConfig type: object description: Configuration for Soniox streaming STT models. required: false properties: - name: languageHints type: array description: >- Language hints to guide the model. If set, will override the `language` field from the main config. required: false properties: - name: item type: string required: false - name: languageHintsStrict type: boolean description: >- If true, model will strongly prefer producing languages only from the `languageHints` list. required: false - name: enableEndpointDetection type: boolean description: >- If true, enables intelligent semantic-based end-of-turn detection. required: false - name: maxEndpointDelayMs type: integer description: >- Maximum allowed delay between the end of the previous turn and the start of the next turn (milliseconds). Must be between 500 and 5000 milliseconds, default is 2000. required: false - name: context type: object description: Contextual information to guide the Soniox model. required: false properties: - name: general type: object description: >- Structured key-value information (domain, topic, intent, etc.). required: false - name: text type: string description: Longer free-form background text or related documents. required: false - name: terms type: array description: Domain-specific or uncommon words. required: false properties: - name: item type: string required: false - name: voiceProfileConfig type: object description: Configuration for voice profile detection. required: false properties: - name: enableVoiceProfile type: boolean description: Enables voice profile feature for this request or stream. required: true - name: topN type: integer description: >- Number of top labels from each class to return. Default: 10. required: false headers: [] jsonPayloadSchema: type: object properties: modelId: type: string description: >- The identifier of the model to use for transcription. Format: "{provider}/{model-name}". Available models: - `inworld/inworld-stt-1` — Inworld first-party - `assemblyai/universal-streaming-multilingual` — AssemblyAI multilingual - `assemblyai/universal-streaming-english` — AssemblyAI English - `assemblyai/u3-rt-pro` — AssemblyAI high-accuracy - `assemblyai/whisper-rt` — AssemblyAI Whisper real-time - `soniox/stt-rt-v4` — Soniox real-time - `soniox/stt-rt-v5` — Soniox real-time (latest generation) - `deepgram/flux-general-en` — Deepgram Flux English conversational - `deepgram/flux-general-multi` — Deepgram Flux multilingual conversational See [STT Introduction](/stt/overview) for the full model catalogue. example: assemblyai/universal-streaming-multilingual x-parser-schema-id: audioEncoding: type: string enum: - AUDIO_ENCODING_UNSPECIFIED - AUTO_DETECT - LINEAR16 - MP3 - OGG_OPUS - FLAC default: AUDIO_ENCODING_UNSPECIFIED description: |- Supported audio encoding formats. - `AUDIO_ENCODING_UNSPECIFIED`: Not specified. Will return an error. - `AUTO_DETECT`: Automatically detect audio encoding from the audio header. - `LINEAR16`: Uncompressed 16-bit signed little-endian samples (Linear PCM). - `MP3`: MP3 audio. Not supported for streaming transcription. - `OGG_OPUS`: Opus encoded audio wrapped in an OGG container. Not supported for streaming transcription. - `FLAC`: FLAC encoded audio. Lossless format. Not supported for streaming transcription. x-parser-schema-id: AudioEncoding language: type: string description: >- Language hint in ISO 639 format (e.g., "en", "ja"). Biases the model toward the specified language during automatic language detection. BCP-47 codes (e.g., "en-US") are also accepted and converted to the base language code. For the Inworld first-party model (`inworld/inworld-stt-1`), the hint additionally constrains the output script for English, Chinese, Cantonese, Japanese, Korean, Russian, and Hindi (e.g. selecting `en` keeps output in Latin script). See [Language Support](/stt/languages) for the full list of supported languages. example: en x-parser-schema-id: sampleRateHertz: type: integer format: int32 description: >- Sample rate of the audio data in Hertz. Required when the sample rate cannot be inferred from the audio header (e.g., raw PCM streams). Default: 16000. example: 16000 x-parser-schema-id: numberOfChannels: type: integer format: int32 description: >- Number of channels in the audio data. Required when the number of channels cannot be inferred from the audio header (e.g., raw PCM streams). Default: 1. example: 1 x-parser-schema-id: inactivityTimeoutSeconds: type: integer format: int32 description: >- Inactivity timeout in seconds. If the client is silent for this duration, the transcription will be stopped. x-parser-schema-id: endOfTurnConfidenceThreshold: type: number format: float description: >- Confidence threshold for end-of-turn prediction. Higher values reduce false-positives. Range: [0.0, 1.0]. Default: 0.5. See the [Turn Detection guide](/stt/turn-detection) for tuning guidance. x-parser-schema-id: prompts: type: array items: type: string x-parser-schema-id: description: >- Custom vocabulary / key terms. An array of context strings (names, jargon, acronyms) that bias the model toward recognizing these terms. This is a soft bias that helps with ambiguous or uncommon words; it is not a hard keyword lock and does not force exact output. Supported across models — the unified field maps to each provider's mechanism (Groq `prompt`, AssemblyAI `keyterms_prompt`, Soniox `context`). Use letters, digits, spaces, and basic punctuation; other characters (such as #, /, @, or |) are rejected by the gateway with INVALID_ARGUMENT (code 3). x-parser-schema-id: includeWordTimestamps: type: boolean description: >- If true, includes per-word timing information in the response. Available for AssemblyAI and Soniox models. Not yet supported for `inworld/inworld-stt-1`. x-parser-schema-id: groqConfig: type: object properties: temperature: type: number format: float description: >- Temperature for the model. Controls randomness in predictions. Higher values produce more varied output. Range: [0.0, 1.0]. x-parser-schema-id: description: Configuration for Groq streaming STT models. x-parser-schema-id: GroqConfig assemblyaiConfig: type: object properties: minEndOfTurnSilenceWhenConfident: type: integer format: int32 description: >- Minimum silence duration when confidence is high (milliseconds). x-parser-schema-id: maxTurnSilence: type: integer format: int32 description: >- Maximum allowed silence before forcing a turn boundary (milliseconds). x-parser-schema-id: vadThreshold: type: number format: float description: >- Voice activity detection threshold. Range: [0.0, 1.0]. Default: 0.5. x-parser-schema-id: prompt: type: string description: >- Custom transcription instructions for the model. Works only for Universal-3 Pro Streaming. x-parser-schema-id: description: Configuration for AssemblyAI streaming STT models. x-parser-schema-id: AssemblyAIConfig inworldSttV1Config: type: object properties: minEndOfTurnSilenceWhenConfident: type: integer format: int32 description: >- Minimum silence duration when confidence is high (milliseconds). x-parser-schema-id: vadThreshold: type: number format: float description: >- Voice activity detection threshold. Range: [0.0, 1.0]. Default: 0.5. x-parser-schema-id: description: >- Configuration for Inworld STT 1 models. Set `vadThreshold` to 0 to disable server-side turn detection and control turn boundaries manually via `endTurn` — see the [Turn Detection guide](/stt/turn-detection). x-parser-schema-id: InworldSttV1Config sonioxConfig: type: object properties: languageHints: type: array items: type: string x-parser-schema-id: description: >- Language hints to guide the model. If set, will override the `language` field from the main config. x-parser-schema-id: languageHintsStrict: type: boolean description: >- If true, model will strongly prefer producing languages only from the `languageHints` list. x-parser-schema-id: enableEndpointDetection: type: boolean description: >- If true, enables intelligent semantic-based end-of-turn detection. x-parser-schema-id: maxEndpointDelayMs: type: integer format: int32 description: >- Maximum allowed delay between the end of the previous turn and the start of the next turn (milliseconds). Must be between 500 and 5000 milliseconds, default is 2000. x-parser-schema-id: context: type: object properties: general: type: object additionalProperties: type: string x-parser-schema-id: description: >- Structured key-value information (domain, topic, intent, etc.). x-parser-schema-id: text: type: string description: Longer free-form background text or related documents. x-parser-schema-id: terms: type: array items: type: string x-parser-schema-id: description: Domain-specific or uncommon words. x-parser-schema-id: description: Contextual information to guide the Soniox model. x-parser-schema-id: SonioxConfigContext description: Configuration for Soniox streaming STT models. x-parser-schema-id: SonioxConfig voiceProfileConfig: type: object properties: enableVoiceProfile: type: boolean description: Enables voice profile feature for this request or stream. x-parser-schema-id: topN: type: integer format: int32 description: 'Number of top labels from each class to return. Default: 10.' x-parser-schema-id: required: - enableVoiceProfile description: Configuration for voice profile detection. x-parser-schema-id: VoiceProfileConfig required: - modelId - audioEncoding examples: - modelId: assemblyai/universal-streaming-multilingual audioEncoding: LINEAR16 sampleRateHertz: 16000 language: en-US x-parser-schema-id: TranscribeConfigPayload title: Transcribe config description: >- Configure the transcription session. Must be the first message sent. Contains model selection, audio format settings, and optional feature configurations. example: |- { "modelId": "assemblyai/universal-streaming-multilingual", "audioEncoding": "LINEAR16", "sampleRateHertz": 16000, "language": "en-US" } bindings: [] extensions: - id: x-parser-unique-object-id value: transcribeConfig - &ref_4 id: audioChunk payload: - name: audioChunk description: >- Send a chunk of audio data for transcription. Must be sent after the initial transcribe config message. type: object properties: - name: content type: string description: >- The raw audio bytes in the encoding specified by the transcribe config's audioEncoding. required: true headers: [] jsonPayloadSchema: type: object properties: content: type: string format: byte description: >- The raw audio bytes in the encoding specified by the transcribe config's audioEncoding. example: x-parser-schema-id: required: - content examples: - content: x-parser-schema-id: AudioChunkPayload title: Audio chunk description: >- Send a chunk of audio data for transcription. Must be sent after the initial transcribe config message. example: |- { "content": "" } bindings: [] extensions: - id: x-parser-unique-object-id value: audioChunk - &ref_5 id: endTurn payload: - name: endTurn description: >- Signal the end of a speaker's turn. Some providers do not support manual turn-taking; for those providers, sending this message will have no effect. See the [Turn Detection guide](/stt/turn-detection) for automatic vs. manual turn control. type: object properties: [] headers: [] jsonPayloadSchema: type: object properties: {} examples: - {} x-parser-schema-id: EndTurnPayload title: End turn description: >- Signal the end of a speaker's turn. Some providers do not support manual turn-taking; for those providers, sending this message will have no effect. See the [Turn Detection guide](/stt/turn-detection) for automatic vs. manual turn control. example: '{}' bindings: [] extensions: - id: x-parser-unique-object-id value: endTurn - &ref_6 id: closeStream payload: - name: closeStream description: >- Signal that the client is done sending audio data. Required for HTTP/WebSocket clients since there is no equivalent to gRPC stream close. type: object properties: [] headers: [] jsonPayloadSchema: type: object properties: {} examples: - {} x-parser-schema-id: CloseStreamPayload title: Close stream description: >- Signal that the client is done sending audio data. Required for HTTP/WebSocket clients since there is no equivalent to gRPC stream close. example: '{}' bindings: [] extensions: - id: x-parser-unique-object-id value: closeStream bindings: [] extensions: &ref_0 - id: x-parser-unique-object-id value: sttStream - &ref_2 id: receiveResponse title: Receive response type: send messages: - &ref_7 id: transcription payload: - name: transcription description: >- Transcription result streamed back as audio is processed. May be an interim (partial) result or a final result depending on the `isFinal` field. type: object properties: - name: transcript type: string description: Full transcribed text for this segment. required: false - name: isFinal type: boolean description: >- Indicates whether this is a finalized result or an interim (partial) result that may be updated as more audio is processed. required: false - name: wordTimestamps type: array description: >- Per-word timing and confidence data. Only populated when `includeWordTimestamps` is enabled. Available for AssemblyAI and Soniox models. Not yet supported for `inworld/inworld-stt-1`. required: false properties: - name: word type: string description: The transcribed word. required: false - name: confidence type: number description: >- Recognition confidence score for this word. Range: [0.0, 1.0]. required: false - name: startTimeMs type: integer description: >- Offset from the beginning of the audio to the start of this word, in milliseconds. required: false - name: endTimeMs type: integer description: >- Offset from the beginning of the audio to the end of this word, in milliseconds. required: false headers: [] jsonPayloadSchema: type: object properties: transcript: type: string description: Full transcribed text for this segment. x-parser-schema-id: isFinal: type: boolean description: >- Indicates whether this is a finalized result or an interim (partial) result that may be updated as more audio is processed. x-parser-schema-id: wordTimestamps: type: array items: type: object properties: word: type: string description: The transcribed word. x-parser-schema-id: confidence: type: number format: float description: >- Recognition confidence score for this word. Range: [0.0, 1.0]. x-parser-schema-id: startTimeMs: type: integer format: int32 description: >- Offset from the beginning of the audio to the start of this word, in milliseconds. x-parser-schema-id: endTimeMs: type: integer format: int32 description: >- Offset from the beginning of the audio to the end of this word, in milliseconds. x-parser-schema-id: x-parser-schema-id: WordTimestamp description: >- Per-word timing and confidence data. Only populated when `includeWordTimestamps` is enabled. Available for AssemblyAI and Soniox models. Not yet supported for `inworld/inworld-stt-1`. x-parser-schema-id: examples: - transcript: Hello, this is a test transcription. isFinal: true wordTimestamps: [] - transcript: Hello, this is isFinal: false wordTimestamps: [] x-parser-schema-id: TranscriptionResponsePayload title: Transcription description: >- Transcription result streamed back as audio is processed. May be an interim (partial) result or a final result depending on the `isFinal` field. example: |- { "transcript": "Hello, this is a test transcription.", "isFinal": true, "wordTimestamps": [] } bindings: [] extensions: - id: x-parser-unique-object-id value: transcription - &ref_8 id: usage payload: - name: usage description: >- Usage metrics for billing and monitoring purposes. **Coming soon** — this field is not yet populated. type: object properties: - name: transcribedAudioMs type: integer description: The duration of the transcribed audio in milliseconds. required: false - name: modelId type: string description: The identifier of the model used for transcription. required: false headers: [] jsonPayloadSchema: type: object properties: transcribedAudioMs: type: integer format: int32 description: The duration of the transcribed audio in milliseconds. x-parser-schema-id: modelId: type: string description: The identifier of the model used for transcription. x-parser-schema-id: description: >- Usage metrics for billing and monitoring purposes. **Coming soon** — this field is not yet populated. x-parser-schema-id: UsageResponsePayload title: Usage description: >- Usage metrics for billing and monitoring purposes. **Coming soon** — this field is not yet populated. example: |- { "transcribedAudioMs": 123, "modelId": "" } bindings: [] extensions: - id: x-parser-unique-object-id value: usage - &ref_9 id: speechStarted payload: - name: speechStarted description: >- Signal to indicate the start of a speaker's speech. Sent when voice activity is detected in the audio stream. type: object properties: - name: startTimeMs type: integer description: The timestamp of the start of the speech in milliseconds. required: false - name: confidence type: number description: The confidence score of the speech detection. required: false headers: [] jsonPayloadSchema: type: object properties: startTimeMs: type: integer format: int32 description: The timestamp of the start of the speech in milliseconds. x-parser-schema-id: confidence: type: number format: float description: The confidence score of the speech detection. x-parser-schema-id: examples: - startTimeMs: 1250 confidence: 0.95 x-parser-schema-id: SpeechStartedResponsePayload title: Speech started description: >- Signal to indicate the start of a speaker's speech. Sent when voice activity is detected in the audio stream. example: |- { "startTimeMs": 1250, "confidence": 0.95 } bindings: [] extensions: - id: x-parser-unique-object-id value: speechStarted - &ref_10 id: speechStopped payload: - name: speechStopped description: >- Signal raised when STT detects silence after speech has stopped. Useful for tracking pauses and implementing custom turn-taking logic. type: object properties: - name: silenceDurationMs type: integer description: >- The duration of silence detected after speech stopped, in milliseconds. required: false headers: [] jsonPayloadSchema: type: object properties: silenceDurationMs: type: integer format: int32 description: >- The duration of silence detected after speech stopped, in milliseconds. x-parser-schema-id: examples: - silenceDurationMs: 750 x-parser-schema-id: SpeechStoppedResponsePayload title: Speech stopped description: >- Signal raised when STT detects silence after speech has stopped. Useful for tracking pauses and implementing custom turn-taking logic. example: |- { "silenceDurationMs": 750 } bindings: [] extensions: - id: x-parser-unique-object-id value: speechStopped bindings: [] extensions: *ref_0 sendOperations: - *ref_1 receiveOperations: - *ref_2 sendMessages: - *ref_3 - *ref_4 - *ref_5 - *ref_6 receiveMessages: - *ref_7 - *ref_8 - *ref_9 - *ref_10 extensions: - id: x-parser-unique-object-id value: sttStream securitySchemes: - id: auth name: authorization type: httpApiKey description: >- Your [authentication](../../../api-reference/introduction) credentials. For Basic authentication, please populate `Basic $INWORLD_API_KEY`. You can create a key in one command with the [Inworld CLI](../../../tts/resources/inworld-cli): `inworld workspace add-key`. in: query extensions: [] ````