Audiovisual processing

Explicit bounded sampling only. No face or voice identity matching, public label changes, new transcription or automatic start. Keep the worker stopped between tests/meetings.

Saved session ID Source Recording start (video seconds) Clip duration (seconds) Spacing / live sampling interval (seconds) Maximum windows (explicit compute scope) Live video time = capture time + offset Anonymous seat/view regions JSON (optional) I checked source A/V and transcript alignment for these intervals

HLS requires matching UTC program-date-time on both audio and video playlists; missing or discontinuous clocks pause this producer. A separately fetched audio transcript can still drift. Recording windows run sequentially; live windows never accumulate a queue. Camera cuts and ambiguity remain unverified. Native Chrome sampling is configured on Capture instead.