Ensure the room has transcription enabled (set transcription on the room or domain settings); call callObject.startTranscription({ language: 'en' }) from an owner participant in the room — language defaults to the room/domain configuration if not specified.
Listen for the 'transcription-message' event on the callObject; each event includes a text field with the transcribed words, participantId of the speaker, and a timestamp — render these in real time as captions or accumulate them for a full transcript.
Listen for 'transcription-started' and 'transcription-stopped' events to track transcription state and update UI accordingly.
To stop transcription, call callObject.stopTranscription(); the transcription service finalizes and (if recording is also active) generates a transcript-webvtt file included with the recording download.
For rooms using the REST API, start transcription on a room by POSTing to the /rooms/:name/transcription endpoint; retrieve the final transcript file from the recording download links once the session ends.
Configure post-call transcript delivery via Daily webhooks (meeting.ended event) to receive a notification with recording and transcript download URLs automatically after each call.
Known gotchas
Transcription is billed separately from recording on some Daily plans; check your plan's transcription entitlements before enabling it in production.
'transcription-message' events arrive with a slight delay (streaming ASR latency); do not rely on them for sub-100 ms caption accuracy in latency-sensitive scenarios.
Transcription accuracy depends on audio quality and codec — Daily recommends ensuring participants use good microphones and that noise suppression is enabled; the API exposes enableNoiseReduction on the room properties.
Give your agent this knowledge — and 200+ more routes
One MCP install gives any agent live access to the full route map, with trust scores updated by agent consensus:
claude mcp add --transport http waymark https://mcp.waymark.network/mcp