Timeline

Likely speech by default; use “Low speech” to review false triggers (keyboard, clicks)

Loading recordings...

Workers

Connected audio capture nodes

Register a New Capture Worker

Use the TranscribeCluster agent on the PC (MicRecordBot enabled in micrecordbot.json) or call POST /api/workers/register. The server returns a token to store in micrecordbot.json on that machine.

Note: Home-page GPU machines are transcribe workers; this list is audio capture nodes for MicRecordBot.

Settings

Recording, storage, and transcription configuration

Storage
Where audio clips are saved. Use /data/nvrbot/clips (Docker volume) or any /mnt path.
Clips older than this are automatically deleted.
Voice Activity Detection
2 — Balanced
0 = most permissive, 3 = most aggressive (loud clear speech only).
Audio captured before speech is detected.
Audio captured after speech ends before saving clip.
Clips shorter than this are discarded.
Remote capture (all machines)
When unchecked, agents stop uploading clips until turned on again (same flag as the main TranscribeCluster "Pause all MicRecordBot" button).
Uses the timezone below. Overnight windows supported (e.g. 22:00–06:00).
Examples: UTC, America/Chicago, Europe/London.
Transcription
Used when Transcribe URL is empty (clips go to your connected GPU workers). Larger models need more VRAM on the worker.
Passed to the same worker pipeline as normal video jobs.
Leave empty (default): clips use the same GPU workers as the main site, but jobs are not listed on the TranscribeCluster home page — status and text appear only here. Set a base URL only if you want HTTP POST …/transcribe on another host instead.
Maintenance
Immediately delete all clips older than the retention period.