> ## 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 speech to text over WebSocket > AsyncAPI specification for the Telnyx Speech-to-Text WebSocket endpoint. Real-time speech transcription by streaming audio and receiving transcript frames. ## Supported Engines - `Azure` - Microsoft Azure Speech Services - `Deepgram` - Deepgram Nova models - `Google` - Google Cloud Speech-to-Text - `Telnyx` - Telnyx native transcription (OpenAI Whisper models) - `xAI` - xAI Grok STT - `AssemblyAI` - AssemblyAI Universal-Streaming - `Speechmatics` - Speechmatics real-time transcription - `Soniox` - Soniox real-time transcription - `Parakeet` - Self-hosted NVIDIA Parakeet multilingual transcription - `Reson8` - Reson8 turn-based multilingual transcription ## Connection Flow 1. Open WebSocket connection to `wss://api.telnyx.com/v2/speech-to-text/transcription` with query parameters. 2. Send binary audio frames (mp3 or wav format). 3. Receive JSON transcript frames with `transcript`, `is_final`, and `confidence` fields. 4. Close connection when done. ## Authentication Requires authentication via a Bearer token (Telnyx API v2 key). ## AsyncAPI ````yaml openapi/asyncapi/speech-to-text-websockets/speech-to-text-websockets.yml speech-to-text-websockets id: speech-to-text-websockets title: Stream speech to text over WebSocket description: > AsyncAPI specification for the Telnyx Speech-to-Text WebSocket endpoint. Real-time speech transcription by streaming audio and receiving transcript frames. ## Supported Engines - `Azure` - Microsoft Azure Speech Services - `Deepgram` - Deepgram Nova models - `Google` - Google Cloud Speech-to-Text - `Telnyx` - Telnyx native transcription (OpenAI Whisper models) - `xAI` - xAI Grok STT - `AssemblyAI` - AssemblyAI Universal-Streaming - `Speechmatics` - Speechmatics real-time transcription - `Soniox` - Soniox real-time transcription - `Parakeet` - Self-hosted NVIDIA Parakeet multilingual transcription - `Reson8` - Reson8 turn-based multilingual transcription ## Connection Flow 1. Open WebSocket connection to `wss://api.telnyx.com/v2/speech-to-text/transcription` with query parameters. 2. Send binary audio frames (mp3 or wav format). 3. Receive JSON transcript frames with `transcript`, `is_final`, and `confidence` fields. 4. Close connection when done. ## Authentication Requires authentication via a Bearer token (Telnyx API v2 key). servers: - id: speech-to-text-websockets protocol: wss host: api.telnyx.com bindings: - protocol: ws version: 0.1.0 value: {} schemaProperties: [] variables: [] address: /v2/speech-to-text/transcription parameters: [] bindings: - protocol: ws version: 0.1.0 value: query: type: object description: Query parameters passed when opening the WebSocket connection. required: - transcription_engine - input_format properties: transcription_engine: type: string description: The transcription engine to use for processing the audio stream. enum: - Azure - Deepgram - Google - Telnyx - xAI - AssemblyAI - Speechmatics - Soniox - Parakeet - Reson8 input_format: type: string description: The format of input audio stream. enum: - mp3 - wav language: type: string description: | The language spoken in the audio stream (e.g., en-US, es-ES). interim_results: type: boolean description: >- Whether to receive interim transcription results before final results. default: false model: type: string description: > The specific model to use within the selected transcription engine. Available models vary by engine: - Azure: `fast` - Deepgram: `deepgram/nova-2`, `deepgram/nova-3` - Google: `latest_long`, `latest_short`, `command_and_search`, `phone_call`, `video`, `default`, `medical_conversation`, `medical_dictation` - Telnyx: `openai/whisper-tiny`, `openai/whisper-large-v3-turbo` - xAI: `xai/grok-stt` - AssemblyAI: `assemblyai/universal-streaming` - Speechmatics: `speechmatics/standard` - Soniox: `soniox/stt-rt-v4` - Parakeet: `nvidia/parakeet-v3` - Reson8: `reson8/turns` endpointing: type: integer description: > Silence duration (in milliseconds) that triggers end-of-speech detection. When set, the engine uses this value to determine when a speaker has stopped talking. Supported by Deepgram, Google, xAI, Speechmatics, and Soniox. Other engines ignore this parameter. For Soniox, the value is interpreted as `max_endpoint_delay_ms` (range: 500–3000 ms). redact: type: string description: > Enable redaction of sensitive information (e.g., PCI data, SSN) from transcription results. Supported values depend on the transcription engine. keyterm: type: string description: | A key term to boost in the transcription. The engine will be more likely to recognize this term. Can be specified multiple times for multiple terms. keywords: type: string description: | Comma-separated list of keywords to boost in the transcription. The engine will prioritize recognition of these words. schemaProperties: - name: query type: object description: Query parameters passed when opening the WebSocket connection. required: false properties: - name: transcription_engine type: string description: The transcription engine to use for processing the audio stream. enumValues: - Azure - Deepgram - Google - Telnyx - xAI - AssemblyAI - Speechmatics - Soniox - Parakeet - Reson8 required: true - name: input_format type: string description: The format of input audio stream. enumValues: - mp3 - wav required: true - name: language type: string description: | The language spoken in the audio stream (e.g., en-US, es-ES). required: false - name: interim_results type: boolean description: >- Whether to receive interim transcription results before final results. required: false - name: model type: string description: > The specific model to use within the selected transcription engine. Available models vary by engine: - Azure: `fast` - Deepgram: `deepgram/nova-2`, `deepgram/nova-3` - Google: `latest_long`, `latest_short`, `command_and_search`, `phone_call`, `video`, `default`, `medical_conversation`, `medical_dictation` - Telnyx: `openai/whisper-tiny`, `openai/whisper-large-v3-turbo` - xAI: `xai/grok-stt` - AssemblyAI: `assemblyai/universal-streaming` - Speechmatics: `speechmatics/standard` - Soniox: `soniox/stt-rt-v4` - Parakeet: `nvidia/parakeet-v3` - Reson8: `reson8/turns` required: false - name: endpointing type: integer description: > Silence duration (in milliseconds) that triggers end-of-speech detection. When set, the engine uses this value to determine when a speaker has stopped talking. Supported by Deepgram, Google, xAI, Speechmatics, and Soniox. Other engines ignore this parameter. For Soniox, the value is interpreted as `max_endpoint_delay_ms` (range: 500–3000 ms). required: false - name: redact type: string description: > Enable redaction of sensitive information (e.g., PCI data, SSN) from transcription results. Supported values depend on the transcription engine. required: false - name: keyterm type: string description: | A key term to boost in the transcription. The engine will be more likely to recognize this term. Can be specified multiple times for multiple terms. required: false - name: keywords type: string description: | Comma-separated list of keywords to boost in the transcription. The engine will prioritize recognition of these words. required: false operations: - &ref_1 id: sendAudio title: Send Audio to Transcribe description: Client sends binary audio frames to be transcribed into text. type: receive messages: - &ref_3 id: clientAudio contentType: application/octet-stream payload: - type: string format: binary description: > Client-to-server binary frame containing audio data to transcribe. Audio should be in mp3 or wav format as specified in the input_format query parameter. x-parser-schema-id: name: Audio Frame headers: [] jsonPayloadSchema: type: string format: binary description: Binary audio data in mp3 or wav format. x-parser-schema-id: title: Audio Frame description: > Client-to-server binary frame containing audio data to transcribe. Audio should be in mp3 or wav format as specified in the input_format query parameter. example: '{}' bindings: [] extensions: - id: x-parser-unique-object-id value: clientAudio bindings: [] extensions: &ref_0 - id: x-parser-unique-object-id value: speech-to-text-websockets - &ref_2 id: receiveTranscript title: Receive Transcript Frames description: Server sends transcript frames and error frames to the client. type: send messages: - &ref_4 id: serverTranscript contentType: application/json payload: - name: Transcript Frame description: > Server-to-client frame containing a transcription result. When interim_results is enabled, you may receive multiple interim results (is_final=false) before the final result (is_final=true) for each utterance. Reson8 transcribes per turn of speech: each interim frame carries the full transcript of the turn so far and supersedes the previous frame, the final frame arrives when the turn ends, and confidence is omitted from its frames. type: object properties: - name: type type: string description: Frame type identifier. required: true - name: transcript type: string description: The transcribed text from the audio. required: true - name: is_final type: boolean description: | Whether this is a final transcription result. When false, this is an interim result that may be refined. required: false - name: confidence type: number description: Confidence score of the transcription, ranging from 0 to 1. required: false headers: [] jsonPayloadSchema: type: object required: - type - transcript properties: type: type: string const: transcript description: Frame type identifier. x-parser-schema-id: transcript: type: string description: The transcribed text from the audio. x-parser-schema-id: is_final: type: boolean description: | Whether this is a final transcription result. When false, this is an interim result that may be refined. x-parser-schema-id: confidence: type: number description: Confidence score of the transcription, ranging from 0 to 1. minimum: 0 maximum: 1 x-parser-schema-id: x-parser-schema-id: title: Transcript Frame description: > Server-to-client frame containing a transcription result. When interim_results is enabled, you may receive multiple interim results (is_final=false) before the final result (is_final=true) for each utterance. Reson8 transcribes per turn of speech: each interim frame carries the full transcript of the turn so far and supersedes the previous frame, the final frame arrives when the turn ends, and confidence is omitted from its frames. example: |- { "type": "transcript", "transcript": "Hello, this is", "is_final": false, "confidence": 0.85 } bindings: [] extensions: - id: x-parser-unique-object-id value: serverTranscript - &ref_5 id: serverError contentType: application/json payload: - name: Error Frame description: | Server-to-client frame indicating an error during transcription. The connection may be closed shortly after sending this frame. type: object properties: - name: type type: string description: Frame type identifier. required: true - name: error type: string description: Error message describing what went wrong. required: true headers: [] jsonPayloadSchema: type: object required: - type - error properties: type: type: string const: error description: Frame type identifier. x-parser-schema-id: error: type: string description: Error message describing what went wrong. x-parser-schema-id: x-parser-schema-id: title: Error Frame description: | Server-to-client frame indicating an error during transcription. The connection may be closed shortly after sending this frame. example: |- { "type": "error", "error": "Invalid transcription_engine specified" } bindings: [] extensions: - id: x-parser-unique-object-id value: serverError bindings: [] extensions: *ref_0 sendOperations: - *ref_1 receiveOperations: - *ref_2 sendMessages: - *ref_3 receiveMessages: - *ref_4 - *ref_5 extensions: - id: x-parser-unique-object-id value: speech-to-text-websockets securitySchemes: - id: bearerAuth name: bearerAuth type: http description: Telnyx API v2 Bearer token authentication. scheme: bearer extensions: [] ````