Docs/Live Transcription

Live Transcription

Stream microphone audio directly to AmiVoice via WebSocket for real-time speech-to-text in the browser.

⚠️Requires Live Transcription Pack. An admin must enable this feature pack in Admin → Features before users can access the Live page.

How it works

Live transcription runs entirely in the browser. The backend only provides connection credentials — audio data is sent directly from the browser to AmiVoice.

Browser → GET /api/live-transcription/config → fetch WS URL + API key
Browser → getUserMedia → AudioContext(16kHz) → AudioWorklet → Int16 PCM → WS binary frames → AmiVoice
  1. Browser fetches AmiVoice WS URL and API key from GET /api/live-transcription/config.
  2. getUserMedia requests microphone access.
  3. AudioContext initialised at 16 kHz; inline AudioWorklet converts Float32 to Int16.
  4. Binary frames (prefixed with byte 0x70) are sent to AmiVoice via WebSocket.
  5. AmiVoice responses are rendered on screen in real time.

WebSocket endpoint

Default endpoint: wss://acp-api.amivoice.com/v1/nolog/

Override with the AMIVOICE_WS_URL environment variable (e.g. to switch to the logging endpoint).

Response events

EventMeaning
UInterim result — not yet finalised
A / RFinal result — confirmed recognition
s Session error (note trailing space)
eSession closed

Requirements and gotchas

  • HTTPS required: navigator.mediaDevices is only available on HTTPS or localhost.
  • Microphone permission: The browser will prompt for microphone access.
  • Language selection: Select the language (ja-JP / en-US) in the UI before starting.
  • Not auto-saved: Live transcription is for real-time monitoring. Results are not automatically saved as voicelog records.
ℹ️Do not include sampleRate in getUserMedia constraints — some audio drivers reject it with NotFoundError.

How to use live transcription

  1. Click Live in the navigation bar (visible only when the feature is enabled).
  2. Select a language (default: ja-JP).
  3. Click Start Recording and grant microphone permission.
  4. Speak — results appear on screen in real time.
  5. Click Stop Recording to end the session.

Next steps

← All DocsContact Support →