Events Lanson Live API pushes over WS /v1/audio/transcriptions/stream. Event identifiers remain unchanged by product or technology branding.
StableStream™ and LiveFinal™ on this endpoint
StableStream™ describes stable, readable live text. LiveFinal™ describes progressive correction and segment finalization. On this public endpoint, optional refinement runs before a single conversation.item.input_audio_transcription.completed event is delivered for each successfully processed utterance.
textis the delivered segment;raw_textis recognition output before optional correction.transcript.stateisstable;transcript.revisionis currently always0.transcript.correction_appliedindicates that correction changed the text. Refinement requires opt-in and gateway availability.- There are no partial/revision events, separate
live.correction.completedevents, or session-wide final transcript events on this endpoint.
Technology names are not wire types. See LiveFinal™ for the processing boundaries and Transcript Lifecycle for client behavior.
Overview
| type | Source | Description |
|---|---|---|
session.created | Gateway | Sent on connection establishment |
session.updated | Gateway | Confirms an applied session.update |
input_audio_buffer.speech_started | Upstream | Speech segment detected |
input_audio_buffer.speech_stopped | Upstream | Speech segment ended |
conversation.item.input_audio_transcription.completed | Upstream | A speech segment has been transcribed (primary output) |
live.translation.completed | Gateway | Translation for one utterance + target language |
lanson.throttled | Gateway | Throttle notification (max once per second) |
error | Gateway / upstream | Error event |
session.created
Sent by the gateway immediately after connection.
{
"type": "session.created",
"request_id": "...",
"session_id": "sess_...",
"audio": { "format": "pcm_s16le", "sample_rate": 16000, "channels": 1 },
"turn_detection": { "type": "server_vad" },
"plan": "pro",
"features": {
"asr_correction": true,
"refinement": false,
"translation": true,
"target_languages": []
},
"limits": {
"max_concurrent_utterances": 8,
"max_session_seconds": 28800,
"idle_timeout_seconds": 180,
"remaining_audio_seconds": 720000
}
}Limits are plan-dependent — the example above shows a pro session. See Rate Limits for all plans.
| Field | Type | Description |
|---|---|---|
session_id | string | Session identifier |
audio.format | string | Fixed: pcm_s16le |
audio.sample_rate | number | Fixed: 16000 |
audio.channels | number | Fixed: 1 |
turn_detection.type | string | Fixed: server_vad |
plan | string | Current plan |
features.asr_correction | boolean | Whether realtime ASR correction is available on this gateway (opt-in per session) |
features.refinement | boolean | string | false when off, or the enabled profile id ("lite" / "pro") |
features.translation | boolean | Whether translation is configured on this gateway |
features.target_languages | string | Frozen translation targets (empty until session.update) |
limits.max_concurrent_utterances | number | Max concurrent utterances |
limits.max_session_seconds | number | Max session duration (seconds) |
limits.idle_timeout_seconds | number | Idle timeout (seconds) |
limits.remaining_audio_seconds | number | null | Remaining audio budget (null when unlimited) |
session.updated
Confirms the state applied by a session.update. Sent for every accepted update, including language-only updates.
{
"type": "session.updated",
"request_id": "...",
"session_id": "sess_...",
"language": "zh",
"language_source": "client",
"target_languages": ["en", "ja"],
"features": {
"asr_correction": true,
"refinement": false,
"translation": true
}
}| Field | Type | Description |
|---|---|---|
language | string | null | Effective source language (null while auto-detection is undecided) |
language_source | string | client or auto |
target_languages | string | Frozen translation targets |
features.asr_correction | boolean | Whether realtime ASR correction is available on this gateway |
features.refinement | boolean | string | false when off, or the enabled profile id ("lite" / "pro") |
features.translation | boolean | true when target_languages is non-empty |
input_audio_buffer.speech_started
VAD detected the start of a speech segment.
{
"type": "input_audio_buffer.speech_started",
"request_id": "...",
"utterance_index": 0
}input_audio_buffer.speech_stopped
VAD detected the end of a speech segment.
{
"type": "input_audio_buffer.speech_stopped",
"request_id": "...",
"utterance_index": 0,
"reason": "end_of_speech",
"audio_duration_ms": 3200,
"speech_duration_ms": 2800
}| Field | Type | Description |
|---|---|---|
utterance_index | number | Speech segment number |
reason | string | Stop reason |
audio_duration_ms | number | Total audio duration |
speech_duration_ms | number? | Actual speech duration |
split_offset_ms | number? | Split offset |
split_rms | number? | Split RMS value |
conversation.item.input_audio_transcription.completed
Primary output event. A speech segment has been transcribed.
{
"type": "conversation.item.input_audio_transcription.completed",
"request_id": "...",
"session_id": "sess_...",
"utterance_index": 0,
"text": "The weather is nice today",
"raw_text": "The weather is nice today",
"language": "en",
"language_source": "auto",
"language_provisional": false,
"audio_duration_ms": 3200,
"latency_ms": 124,
"stt_endpoint": "https://.../v1/audio/transcriptions",
"worker_id": "cohere-s1",
"transcript": {
"id": "tr_0",
"context_id": "ctx_sess_...",
"state": "stable",
"revision": 0,
"safe_for_downstream": true,
"correction_applied": false,
"language_source": "auto",
"language_provisional": false
}
}| Field | Type | Description |
|---|---|---|
utterance_index | number | Speech segment number |
text | string | Final transcribed text |
raw_text | string | Text before optional correction |
language | string | Language used for the request |
language_source | string | client (explicit session.update) or auto (session detection) |
language_provisional | boolean | true while auto-detection has not locked yet (best-effort language) |
language_detection_error | string? | Present when auto-detection failed for this utterance |
audio_duration_ms | number | Audio duration |
latency_ms | number | Processing latency |
stt_endpoint | string | Upstream STT endpoint that produced the result |
worker_id | string | Upstream worker that produced the result |
transcript.id | string | Transcript record id |
transcript.context_id | string | Session-scoped context id |
transcript.state | string | stable for the primary output |
transcript.revision | number | Currently 0; this endpoint does not emit later revisions |
transcript.safe_for_downstream | boolean | Basic readiness signal (false for empty text or a detection error); not an accuracy guarantee |
transcript.correction_applied | boolean | Whether ASR correction changed the text |
transcript.language_source | string | Same as the top-level language_source |
transcript.language_provisional | boolean | Same as the top-level language_provisional |
transcript.language_detection_error | string? | Same as the top-level language_detection_error |
Language fields and UI display
In auto mode, language is the language the session detected and used for this utterance. The client can display it directly:
language_source: "client"— the client explicitly set this language; it is not a detection result.language_source: "auto"— the language came from session detection.language_provisional: true— detection is still voting; show the language as provisional (for example “detecting…” or a muted badge). A later event withlanguage_provisional: falseconfirms the locked language.language_detection_errorpresent — detection failed; show “unknown” and retry on the next utterance rather than displaying a wrong language.
Detection is delivered with the utterance's completed event; there is no separate client event for "language detected". If the UI needs the language before sending audio, call POST /v1/audio/language-detection first.
In auto mode, language_provisional: true means the session is still voting on the language; the text is best-effort. If detection fails, language_detection_error is present and safe_for_downstream is false; the session retries on later speech. Passing an explicit language avoids this state entirely.
live.translation.completed
Translation for one utterance and one target language. One event per target language, emitted independently and never blocking the source transcript.
{
"type": "live.translation.completed",
"request_id": "...",
"session_id": "sess_...",
"utterance_index": 0,
"language": "en",
"text": "The weather is nice today",
"source_text": "今天天气不错",
"source_language": "zh",
"latency_ms": 180
}| Field | Type | Description |
|---|---|---|
utterance_index | number | Matches the source conversation.item.input_audio_transcription.completed event |
language | string | Target language (normalized BCP-47) |
text | string | Translated text |
source_text | string | Raw ASR line the translation is based on |
source_language | string? | Source language used for this utterance |
latency_ms | number | Translation latency |
lanson.throttled
Gateway throttle notification, sent at most once per second.
{
"type": "lanson.throttled",
"reason": "concurrent_utterance_limit",
"dropped_audio_frames": 3,
"in_flight_utterances": 2
}reason value | Meaning |
|---|---|
concurrent_utterance_limit | Concurrent utterance limit exceeded |
client_backpressure | Client is not reading events fast enough |
error
Error event. Can be sent at any time during the WebSocket connection.
{
"type": "error",
"code": "audio_frame_too_large",
"message": "Audio frame exceeds maximum size.",
"request_id": "..."
}Error codes are listed in Errors.
Event ordering
session.created— Session initialized upon WebSocket handshake.session.update— Client optionally setslanguageandtarget_languages(recommended before audio).session.updated— Gateway confirms the applied session state.input_audio_buffer.append— Client streams base64 PCM16 audio frames.input_audio_buffer.speech_started— Server-side VAD detects start of speech.input_audio_buffer.append— Client continues streaming speech audio.input_audio_buffer.speech_stopped— VAD detects silence or segment boundary.conversation.item.input_audio_transcription.completed— Stable transcription text delivered.live.translation.completed— One event per target language, as each translation finishes.- (Repeats for each utterance in the stream)
- Session close.
Related
- Client Messages — client-sent messages
- Errors — full error reference
- Transcript Lifecycle — state transitions
- LiveFinal™ — correction and finalization behavior
On This Page
StableStream™ and LiveFinal™ on this endpointOverviewsession.createdsession.updatedinput_audio_buffer.speech_startedinput_audio_buffer.speech_stoppedconversation.item.input_audio_transcription.completedLanguage fields and UI displaylive.translation.completedlanson.throttlederrorEvent orderingRelated