Realtime Overview

PreviousNext

Stream live speech with Lanson Live API and receive stable, readable text.

Realtime is the live-streaming path of Lanson Live API, LansonAI's Voice Context Layer for Live Speech. Use Segment for client-cut clips or Batch for recordings.

You send PCM16LE audio over WebSocket. Trinity Engine coordinates processing, and the API delivers stable, readable text through StableStream™ behavior. LiveFinal™ describes correction and segment finalization; see its current event boundaries.

What it does

  • Live captions: text appears while the person is still speaking
  • Real-time translation: source and translated text together, per target language
  • Multi-language: Whisper-compatible language support
  • Server-side VAD: automatic speech segment detection, no client-side splitting needed

Input

  • Audio format: PCM16LE / 16kHz / mono
  • Transport: JSON text frames (base64) — the supported transport today
  • Frame size: recommended 100ms frames (~3200 bytes), max 1 MiB

Output

Server pushes JSON events. The primary output is:

  • conversation.item.input_audio_transcription.completed — a speech segment has been transcribed

Supporting events:

  • input_audio_buffer.speech_started / speech_stopped — speech segment boundaries
  • session.created — connection established
  • session.updated — confirmed language and feature settings
  • live.translation.completed — a target-language result when translation is enabled
  • error — error events

The current public endpoint emits one completed result per successfully processed utterance, not partial-text revisions. Optional refinement runs before that result is delivered. It is off by default; the default lite profile is included in the base transcription price, and pro is billed as the realtime refinement add-on. Confirm feature availability and configure it before sending audio.

Architecture

Realtime API Architecture

vs. Segment and Batch

DimensionRealtime WSSegment HTTPBatch jobs
TransportWebSocketHTTP syncHTTP async
LatencyMilliseconds~≤2.4sMinutes
InputPCM16LE streamMultipart fileAudio URL
OutputEvent streamTranscript JSONComplete job JSON
VADServerClientServer (slice)
Best forLive streamPre-cut utterancesLong files

Next steps