mCaptions Media Browser

Each card in the left panel represents one processing job. The coloured badge shows its status (queued / processing / completed / failed / aborted). Three icons indicate whether audio, a transcript, and speaker data are available. Clicking a card opens the viewer on the right.

Active jobs show a thin blue progress bar at the bottom of their card and the percentage inside the badge (e.g. processing 47%). The browser polls in-progress jobs every 3 seconds and detects completion within the same interval. A lightweight structural check (new / deleted jobs) runs every 10 seconds.

Only the 50 most recent jobs are loaded on startup to keep the initial load fast. Click ↓ Load more jobs… at the bottom of the list to fetch the next 50 older jobs. Newly submitted jobs always appear at the top automatically.

Type in the Filter visible jobs… box to narrow the list in real time. The filter matches against job ID, timestamp (e.g. 2026-03), status (e.g. failed), and action type (e.g. summary). The filter is preserved across automatic refreshes.

Use Record to capture audio directly from your microphone (requires HTTPS). Use Upload to submit an existing audio or video file. Both flows let you choose the action (Transcript / Captions / Summarize) and enable speaker diarization before submitting.

Click any speaker chip in the sidebar, or click the coloured speaker tag on a transcript line, to focus on that speaker. All other speakers fade out on the timeline and their transcript lines collapse. Playback skips all other speakers' segments automatically — the playhead flashes gold whenever a gap is jumped. Click the same speaker again to return to normal view.

Upload a .zip of .bin speaker profiles (with a manifest.json) using the Speaker Profiles panel in the sidebar. The browser computes similarity scores between the reference profiles and the detected speakers and labels them automatically. Adjust the Threshold slider (default 0.60) to control the minimum score required for a match — all profiles above the threshold are shown, the best one is applied as the label.