quickstart.md

Plaudio quickstart

Five steps from a fresh install to searchable transcripts.

0. Install and verify

pip install plaudio
plaudio doctor

If doctor reports failures, fix them before continuing.

1. Enrol a voice

You need a ~30-second clip of one person speaking alone. Tell them you're going to record their voice for the bank.

plaudio enrol alice-clean-clip.mp3 --name "Alice Smith" --start 0 --end 30 --num-speakers 1

The --num-speakers 1 hint helps pyannote when the audio really only has one speaker.

2. Transcribe a meeting

plaudio transcribe team-meeting.mp3 --language en

This produces team-meeting.json (raw Whisper output).

3. Convert to Plaudio's transcript shape

Plaudio's .plaud.json shape is:

{
  "meta": {"language": "en", "n_speakers": 3},
  "segments": [
    {"start_time": 0, "end_time": 2000, "speaker": "Speaker 1", "content": "hello"},
    ...
  ]
}

For now you produce this yourself (a short Python script that converts Whisper segments). Full pipeline integration arrives later.

4. Label speakers from the voice bank

plaudio match team-meeting.mp3 team-meeting.plaud.json --threshold 0.55 --report

The report shows per-speaker airtime after labelling. Anyone not in the bank stays as Unknown; enrol them next time.

5. Ingest into the searchable corpus

plaudio db ingest team-meeting.plaud.json \
  --meeting-id 2026-05-30-team \
  --date 2026-05-30 \
  --title "Team weekly"

plaudio db search "release"
plaudio db search "deadline" --speaker "Alice Smith"
plaudio db list