> ## Documentation Index > Fetch the complete documentation index at: https://docs.slng.ai/llms.txt > Use this file to discover all available pages before exploring further. # Unified STT > Stream audio to any WebSocket-capable STT model over SLNG's native protocol. Send `init`, then audio as base64 `audio` messages or raw binary frames; the gateway returns `ready`, `partial_transcript`, `final_transcript`, provider-dependent `utterance_end`, and flat `error` frames. A `final_transcript` is a finalized segment that may be empty and is never a completion signal; `finalize` requests processing while input stays open; `close` is the definitive end of input, after which the client drains remaining results until the gateway closes the WebSocket with code 1000. Send `keepalive` every 3–5 seconds whenever input is open without audio (Nova); a WebSocket ping is not a keepalive. Optional `utterance_final` and `finalize_result` booleans report provider speech endpoints and finalize provenance and must not be treated as acknowledgements. The gateway emits no `duration` or `language` on transcripts. ## AsyncAPI ````yaml api-reference/asyncapi/bridges-unmute.asyncapi.yaml /v1/bridges/unmute/stt/{model_variant} id: /v1/bridges/unmute/stt/{model_variant} title: Unmute STT Bridge description: >- Stream audio to any WebSocket-capable STT model over SLNG's native protocol. Send `init`, then audio as base64 `audio` messages or raw binary frames; the gateway returns `ready`, `partial_transcript`, `final_transcript`, provider-dependent `utterance_end`, and flat `error` frames. A `final_transcript` is a finalized segment that may be empty and is never a completion signal; `finalize` requests processing while input stays open; `close` is the definitive end of input, after which the client drains remaining results until the gateway closes the WebSocket with code 1000. Send `keepalive` every 3–5 seconds whenever input is open without audio (Nova); a WebSocket ping is not a keepalive. Optional `utterance_final` and `finalize_result` booleans report provider speech endpoints and finalize provenance and must not be treated as acknowledgements. The gateway emits no `duration` or `language` on transcripts. servers: - id: production protocol: wss host: api.slng.ai bindings: [] variables: [] address: /v1/bridges/unmute/stt/{model_variant} parameters: [] bindings: - protocol: ws version: latest value: method: GET schemaProperties: - name: method type: string description: GET required: false operations: - &ref_2 id: unmuteSttBridgeReceiveInit title: Unmute stt bridge receive init description: Initialize session on Unmute STT Bridge type: receive messages: - &ref_12 id: UnmuteSttInitMessage contentType: application/json payload: - name: Init (Unmute STT) description: Initialize Unmute STT session with configuration. type: object properties: - name: type type: string description: init required: true - name: config type: object description: Session configuration. required: false properties: - name: language type: string description: ISO-639-1 language code. enumValues: - en - ja - zh - de - hi - fr - ko - pt - it - es - id - nl - tr - fil - pl - sv - bg - ro - ar - cs - el - fi - hr - ms - sk - da - ta - uk - ru - hu - 'no' - vi required: false - name: sample_rate type: integer description: Audio sample rate in Hz. enumValues: - 8000 - 16000 - 22050 - 24000 - 32000 - 44100 - 48000 required: false - name: channels type: integer description: >- Input channel count. Optional on SLNG-hosted Nova routes (defaults to 1); used to meter raw audio exactly. required: false - name: encoding type: string description: >- Input audio encoding. SLNG-hosted Nova routes accept `linear16` (8000/16000 Hz); other targets may accept the remaining values. enumValues: - linear16 - LINEAR16 - pcm_s16le - pcm_l16 - pcm_raw - pcm_f32le - wav - mulaw - mp3 - opus required: false - name: enable_vad type: boolean description: >- Enable voice activity detection where the provider supports it. required: false - name: enable_partials type: boolean description: >- Enable interim `partial_transcript` results. Required for Nova `utterance_end` events. required: false - name: description type: string description: >- Provider-specific (Deepgram) silence detector. Integer milliseconds of silence before a segment is finalized, or `false` to disable. Controls segment finalization, not utterance end. required: false - name: utterance_end_ms type: integer description: >- Provider-specific (Deepgram) word-gap detector in milliseconds that produces `utterance_end` events. Requires `enable_partials: true`. required: false - name: keyterm type: array description: Keyterms to boost recognition on Deepgram Nova 3 models. required: false properties: - name: item type: string required: false headers: [] jsonPayloadSchema: type: object description: >- Optional session configuration sent before audio. Besides the fields below, any provider-specific option declared by the selected model is passed through unchanged (Deepgram-backed models also accept `punctuate`, `smart_format`, `diarize`, `numerals`, `profanity_filter`). On SLNG-hosted Nova HTTP routes `smart_format=true` currently fails upstream while `punctuate=true` works; WebSocket `smart_format` on hosted routes has not been qualified. required: - type properties: type: type: string const: init x-parser-schema-id: config: type: object description: Session configuration. properties: language: type: string description: ISO-639-1 language code. enum: - en - ja - zh - de - hi - fr - ko - pt - it - es - id - nl - tr - fil - pl - sv - bg - ro - ar - cs - el - fi - hr - ms - sk - da - ta - uk - ru - hu - 'no' - vi x-parser-schema-id: LanguageCode sample_rate: type: integer description: Audio sample rate in Hz. enum: - 8000 - 16000 - 22050 - 24000 - 32000 - 44100 - 48000 x-parser-schema-id: SampleRate channels: type: integer minimum: 1 maximum: 16 description: >- Input channel count. Optional on SLNG-hosted Nova routes (defaults to 1); used to meter raw audio exactly. x-parser-schema-id: encoding: type: string description: >- Input audio encoding. SLNG-hosted Nova routes accept `linear16` (8000/16000 Hz); other targets may accept the remaining values. enum: - linear16 - LINEAR16 - pcm_s16le - pcm_l16 - pcm_raw - pcm_f32le - wav - mulaw - mp3 - opus x-parser-schema-id: enable_vad: type: boolean description: >- Enable voice activity detection where the provider supports it. x-parser-schema-id: enable_partials: type: boolean description: >- Enable interim `partial_transcript` results. Required for Nova `utterance_end` events. x-parser-schema-id: endpointing: description: >- Provider-specific (Deepgram) silence detector. Integer milliseconds of silence before a segment is finalized, or `false` to disable. Controls segment finalization, not utterance end. x-parser-schema-id: utterance_end_ms: type: integer minimum: 1000 maximum: 5000 description: >- Provider-specific (Deepgram) word-gap detector in milliseconds that produces `utterance_end` events. Requires `enable_partials: true`. x-parser-schema-id: keyterm: type: array description: Keyterms to boost recognition on Deepgram Nova 3 models. items: type: string x-parser-schema-id: x-parser-schema-id: x-parser-schema-id: x-parser-schema-id: title: Init (Unmute STT) description: Initialize Unmute STT session with configuration. example: |- { "type": "init", "config": { "language": "en", "sample_rate": 16000, "channels": 1, "encoding": "linear16", "enable_partials": true } } bindings: [] extensions: - id: x-parser-unique-object-id value: UnmuteSttInitMessage bindings: [] extensions: &ref_0 - id: x-parser-unique-object-id value: /v1/bridges/unmute/stt/{model_variant} - &ref_3 id: unmuteSttBridgeReceiveAudio title: Unmute stt bridge receive audio description: Send audio to Unmute STT Bridge type: receive messages: - &ref_13 id: UnmuteSttAudioMessage contentType: application/json payload: - name: Audio (Unmute STT) description: Send audio data to Unmute STT Bridge. type: object properties: - name: type type: string description: audio required: true - name: data type: string description: Base64-encoded audio bytes. required: true headers: [] jsonPayloadSchema: type: object description: >- Base64 audio in a JSON text frame. The same bytes may instead be sent as raw binary WebSocket frames; both forms are metered as input audio in the encoding declared by `init`. Silent PCM advances the audio stream and is metered; use `keepalive` for silence instead. required: - type - data properties: type: type: string const: audio x-parser-schema-id: data: type: string format: byte description: Base64-encoded audio bytes. x-parser-schema-id: x-parser-schema-id: title: Audio (Unmute STT) description: Send audio data to Unmute STT Bridge. example: |- { "type": "audio", "data": "UklGRiQAAABXQVZFZm10IBAAAAABAAEA..." } bindings: [] extensions: - id: x-parser-unique-object-id value: UnmuteSttAudioMessage bindings: [] extensions: *ref_0 - &ref_4 id: unmuteSttBridgeReceiveFinalize title: Unmute stt bridge receive finalize description: >- Request processing of buffered audio on Unmute STT Bridge without closing input type: receive messages: - &ref_14 id: UnmuteSttFinalizeMessage contentType: application/json payload: - name: Finalize (Unmute STT) description: Request processing of buffered audio without closing input. type: object properties: - name: type type: string description: finalize required: true headers: [] jsonPayloadSchema: type: object description: >- Requests processing and leaves input open. Nova does not guarantee a response and does not correlate results to this request: zero, one, or several final segments (including empty ones) may follow, and a later result may carry `finalize_result: true` as provenance only. Keep receiving and keep sending keepalives; do not pair finals with finalize requests or implement a completion timeout. Use `close` for a complete input stream. required: - type properties: type: type: string const: finalize x-parser-schema-id: x-parser-schema-id: title: Finalize (Unmute STT) description: Request processing of buffered audio without closing input. example: |- { "type": "finalize" } bindings: [] extensions: - id: x-parser-unique-object-id value: UnmuteSttFinalizeMessage bindings: [] extensions: *ref_0 - &ref_5 id: unmuteSttBridgeReceiveClose title: Unmute stt bridge receive close description: Close audio input on Unmute STT Bridge and drain remaining results type: receive messages: - &ref_15 id: UnmuteSttCloseMessage contentType: application/json payload: - name: Close (Unmute STT) description: Close audio input and drain remaining results. type: object properties: - name: type type: string description: close required: true headers: [] jsonPayloadSchema: type: object description: >- Signals that no more audio will be sent. The gateway forwards the provider's close-stream control, keeps delivering every remaining result, and then closes the WebSocket with code 1000. Successful closure after `close` is the only definitive completion signal; a final transcript or utterance boundary alone is not. Use a new connection for the next independent input. required: - type properties: type: type: string const: close x-parser-schema-id: x-parser-schema-id: title: Close (Unmute STT) description: Close audio input and drain remaining results. example: |- { "type": "close" } bindings: [] extensions: - id: x-parser-unique-object-id value: UnmuteSttCloseMessage bindings: [] extensions: *ref_0 - &ref_6 id: unmuteSttBridgeReceiveKeepalive title: Unmute stt bridge receive keepalive description: Send application keepalive on Unmute STT Bridge type: receive messages: - &ref_16 id: UnmuteSttKeepaliveMessage contentType: application/json payload: - name: Keepalive (Unmute STT) description: Application keepalive while input is open without audio. type: object properties: - name: type type: string description: keepalive required: true headers: [] jsonPayloadSchema: type: object description: >- Carries no audio and produces no reply. For Nova, send every 3–5 seconds whenever input remains open without audio; the provider ends idle streams after roughly ten seconds. The gateway forwards it as the provider's native keepalive and resets its own idle timeout, but it never extends the maximum session duration. A WebSocket ping is answered but is not an application keepalive. required: - type properties: type: type: string const: keepalive x-parser-schema-id: x-parser-schema-id: title: Keepalive (Unmute STT) description: Application keepalive while input is open without audio. example: |- { "type": "keepalive" } bindings: [] extensions: - id: x-parser-unique-object-id value: UnmuteSttKeepaliveMessage bindings: [] extensions: *ref_0 - &ref_7 id: unmuteSttBridgeSendReady title: Unmute stt bridge send ready description: Receive ready signal from Unmute STT Bridge type: send messages: - &ref_17 id: UnmuteSttReadyMessage contentType: application/json payload: - name: Ready (Unmute STT) description: Unmute STT session is ready. type: object properties: - name: type type: string description: ready required: true - name: session_id type: string description: Gateway request identifier for this session. required: true - name: slng_request_id type: string description: >- SLNG request identifier (same value as `session_id`) for log correlation. required: true headers: [] jsonPayloadSchema: type: object description: >- Sent once after the upstream session is open. `session_id` and `slng_request_id` carry the same gateway request identifier; quote it when reporting problems. required: - type - session_id - slng_request_id properties: type: type: string const: ready x-parser-schema-id: session_id: type: string description: Gateway request identifier for this session. x-parser-schema-id: slng_request_id: type: string description: >- SLNG request identifier (same value as `session_id`) for log correlation. x-parser-schema-id: x-parser-schema-id: title: Ready (Unmute STT) description: Unmute STT session is ready. example: |- { "type": "ready", "session_id": "2c9953d7-d4b4-4385-bb27-f64e0d979be5", "slng_request_id": "2c9953d7-d4b4-4385-bb27-f64e0d979be5" } bindings: [] extensions: - id: x-parser-unique-object-id value: UnmuteSttReadyMessage bindings: [] extensions: *ref_0 - &ref_8 id: unmuteSttBridgeSendPartialTranscript title: Unmute stt bridge send partial transcript description: Receive partial transcript from Unmute STT Bridge type: send messages: - &ref_18 id: UnmuteSttPartialTranscript contentType: application/json payload: - name: Partial Transcript (Unmute STT) description: Interim (revisable) transcription result. type: object properties: - name: type type: string description: partial_transcript required: true - name: transcript type: string description: Interim text so far. required: true - name: confidence type: number description: Provider confidence (0-1) when reported. required: false - name: utterance_final type: boolean description: >- Present only when the provider reports a speech endpoint for this result (Nova `speech_final`). Not stream completion and not a turn acknowledgement. required: false - name: finalize_result type: boolean description: >- Present only when the provider attributes this result to finalize processing (Nova `from_finalize`). Informational provenance; uncorrelated with any specific `finalize` and not proof that all submitted audio was processed. required: false headers: [] jsonPayloadSchema: type: object description: >- Revisable segment text; replace provisional text with the latest partial. Only `type` and `transcript` are always present. required: - type - transcript properties: type: type: string const: partial_transcript x-parser-schema-id: transcript: type: string description: Interim text so far. x-parser-schema-id: confidence: type: number description: Provider confidence (0-1) when reported. x-parser-schema-id: utterance_final: type: boolean description: >- Present only when the provider reports a speech endpoint for this result (Nova `speech_final`). Not stream completion and not a turn acknowledgement. x-parser-schema-id: finalize_result: type: boolean description: >- Present only when the provider attributes this result to finalize processing (Nova `from_finalize`). Informational provenance; uncorrelated with any specific `finalize` and not proof that all submitted audio was processed. x-parser-schema-id: x-parser-schema-id: title: Partial Transcript (Unmute STT) description: Interim (revisable) transcription result. example: |- { "type": "partial_transcript", "transcript": "Hello wor", "confidence": 0.85 } bindings: [] extensions: - id: x-parser-unique-object-id value: UnmuteSttPartialTranscript bindings: [] extensions: *ref_0 - &ref_9 id: unmuteSttBridgeSendFinalTranscript title: Unmute stt bridge send final transcript description: Receive a finalized segment (possibly empty) from Unmute STT Bridge type: send messages: - &ref_19 id: UnmuteSttFinalTranscript contentType: application/json payload: - name: Final Transcript (Unmute STT) description: A finalized segment, possibly empty. type: object properties: - name: type type: string description: final_transcript required: true - name: transcript type: string description: Finalized segment text; may be an empty string. required: true - name: confidence type: number description: Provider confidence (0-1) when reported. required: false - name: utterance_final type: boolean description: >- Present only when the provider reports a speech endpoint for this result (Nova `speech_final`). Not stream completion and not a turn acknowledgement. required: false - name: finalize_result type: boolean description: >- Present only when the provider attributes this result to finalize processing (Nova `from_finalize`). Informational provenance; uncorrelated with any specific `finalize` and not proof that all submitted audio was processed. required: false headers: [] jsonPayloadSchema: type: object description: >- A finalized segment. Append nonempty text in arrival order and keep receiving: segment finality does not close input or complete the stream, and an empty segment is never a completion sentinel. Only `type` and `transcript` are always present. The gateway emits no `duration`, `language`, or timing fields on this message; absent optional booleans mean the provider did not report them (they are never filled with `false`). required: - type - transcript properties: type: type: string const: final_transcript x-parser-schema-id: transcript: type: string description: Finalized segment text; may be an empty string. x-parser-schema-id: confidence: type: number description: Provider confidence (0-1) when reported. x-parser-schema-id: utterance_final: type: boolean description: >- Present only when the provider reports a speech endpoint for this result (Nova `speech_final`). Not stream completion and not a turn acknowledgement. x-parser-schema-id: finalize_result: type: boolean description: >- Present only when the provider attributes this result to finalize processing (Nova `from_finalize`). Informational provenance; uncorrelated with any specific `finalize` and not proof that all submitted audio was processed. x-parser-schema-id: x-parser-schema-id: title: Final Transcript (Unmute STT) description: A finalized segment, possibly empty. example: |- { "type": "final_transcript", "transcript": "Hello world", "confidence": 0.95 } bindings: [] extensions: - id: x-parser-unique-object-id value: UnmuteSttFinalTranscript bindings: [] extensions: *ref_0 - &ref_10 id: unmuteSttBridgeSendUtteranceEnd title: Unmute stt bridge send utterance end description: Receive provider-dependent utterance end marker from Unmute STT Bridge type: send messages: - &ref_20 id: UnmuteSttUtteranceEnd contentType: application/json payload: - name: Utterance End (Unmute STT) description: Provider-dependent end-of-utterance marker. type: object properties: - name: type type: string description: utterance_end required: true - name: last_word_end type: number description: >- Audio timestamp in seconds of the last transcribed word, when reported. required: false - name: channel type: array description: Provider channel index for the boundary, when reported. required: false properties: - name: item type: integer required: false - name: stream_id type: integer description: Provider stream identifier (Gradium only). required: false headers: [] jsonPayloadSchema: type: object description: >- Emitted independently of `final_transcript` when the provider reports a word-gap utterance boundary. It is not stream completion and does not cover queued audio. Nova sends it only when `utterance_end_ms` is set together with `enable_partials: true`, with `last_word_end` and `channel`; `stream_id` is supplied by Gradium only. Providers without endpoint detection never send it. required: - type properties: type: type: string const: utterance_end x-parser-schema-id: last_word_end: type: number description: >- Audio timestamp in seconds of the last transcribed word, when reported. x-parser-schema-id: channel: type: array description: Provider channel index for the boundary, when reported. items: type: integer x-parser-schema-id: x-parser-schema-id: stream_id: type: integer description: Provider stream identifier (Gradium only). x-parser-schema-id: x-parser-schema-id: title: Utterance End (Unmute STT) description: Provider-dependent end-of-utterance marker. example: |- { "type": "utterance_end", "last_word_end": 2.395, "channel": [ 0 ] } bindings: [] extensions: - id: x-parser-unique-object-id value: UnmuteSttUtteranceEnd bindings: [] extensions: *ref_0 - &ref_11 id: unmuteSttBridgeSendError title: Unmute stt bridge send error description: Receive error from Unmute STT Bridge type: send messages: - &ref_21 id: UnmuteSttErrorMessage contentType: application/json payload: - name: Error (Unmute STT) description: Error occurred during Unmute STT session. type: object properties: - name: type type: string description: error required: true - name: code type: &ref_1 - string - integer description: >- Error code. Gateway codes are strings; a verbatim provider passthrough keeps the provider's value, which may be numeric. required: true - name: message type: string description: Human-readable error description; wording may change. required: true - name: slng_request_id type: string description: >- SLNG request identifier for log correlation; present on every gateway-generated or gateway-translated error, absent on a verbatim provider passthrough. required: false headers: [] jsonPayloadSchema: type: object description: >- Flat error frame; there is no nested `data` envelope. Gateway-generated codes are `config_error`, `processing_error`, `provider_error`, and `rate_limit` (oversize payloads and message-rate breaches both report `rate_limit`); these and provider errors translated by the gateway always carry `slng_request_id`. A provider frame that already has `type: error` and that the provider's catalog does not declare is forwarded verbatim (reson8), so `code` is then the provider's own value (possibly numeric) and `slng_request_id` is absent. Branch on `code`, not on `message` text. Authentication failures are rejected at the HTTP upgrade and never arrive as this message. required: - type - code - message properties: type: type: string const: error x-parser-schema-id: code: type: *ref_1 description: >- Error code. Gateway codes are strings; a verbatim provider passthrough keeps the provider's value, which may be numeric. x-parser-schema-id: message: type: string description: Human-readable error description; wording may change. x-parser-schema-id: slng_request_id: type: string description: >- SLNG request identifier for log correlation; present on every gateway-generated or gateway-translated error, absent on a verbatim provider passthrough. x-parser-schema-id: x-parser-schema-id: title: Error (Unmute STT) description: Error occurred during Unmute STT session. example: |- { "type": "error", "code": "provider_error", "message": "STT provider completed before input close", "slng_request_id": "2c9953d7-d4b4-4385-bb27-f64e0d979be5" } bindings: [] extensions: - id: x-parser-unique-object-id value: UnmuteSttErrorMessage bindings: [] extensions: *ref_0 sendOperations: - *ref_2 - *ref_3 - *ref_4 - *ref_5 - *ref_6 receiveOperations: - *ref_7 - *ref_8 - *ref_9 - *ref_10 - *ref_11 sendMessages: - *ref_12 - *ref_13 - *ref_14 - *ref_15 - *ref_16 receiveMessages: - *ref_17 - *ref_18 - *ref_19 - *ref_20 - *ref_21 extensions: - id: x-parser-unique-object-id value: /v1/bridges/unmute/stt/{model_variant} securitySchemes: - id: bearer name: bearer type: http description: > API key issued by SLNG. Pass as `Authorization: Bearer ` in the WebSocket upgrade request headers. scheme: bearer extensions: [] ```` This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.