Speaker Diarizer
Answers "who spoke when?" in a meeting or interview recording. Separates speakers automatically and produces a timestamped text report.
What it does
Use this to break a meeting into per-speaker quotes, distinguish the interviewer from the interviewee, or pull individual contributions out of a panel recording.
How to use
- Drag meeting/interview recordings into the list.
- Pick the Expected Speaker Count (Auto-Detect or a number from 2 to 10).
- Pick a Neural Mode: Standard, High Precision, or Turbo.
- Click Run.
You get one timestamped .txt report per file. Format:
[00:00:05] SPEAKER_01: Hi, welcome.
[00:00:12] SPEAKER_02: Thanks.
Speaker count
- Auto-Detect: The system estimates how many speakers there are. Try this first.
- A number from 2 to 10: If you know exactly how many people spoke, this gives better accuracy.
Quality modes
| Mode | What for |
|---|---|
| Standard | Balanced, the everyday choice. |
| High Precision | Important recordings, production use. Slower but more accurate. |
| Turbo | Fast draft, lower accuracy. |
Examples
Break a meeting into speakers: Add the meeting, Auto-Detect, Standard, run. Each speaker is labelled SPEAKER_01, SPEAKER_02 and so on.
Q&A split for an interview: Add the interview, Speakers 2, High Precision, run. Both speakers come out cleanly separated.
Analyse a panel recording: Add the panel, Speakers 4, Standard, run.
Batch interview archive: Add many interviews, Speakers 2, run. Each one gets its own report.
Watch out
- The first run downloads the Whisper model from the internet. Later runs are offline.
- Speakers are labelled SPEAKER_01, SPEAKER_02 - not real names. You can rename them by hand afterwards.
- Very similar voices (siblings, same gender) may get confused.
- Very crowded settings (10+ people) hurt accuracy.
- Music or ambience-heavy recordings produce inconsistent results.
- For a single speaker just run transcription, no need for diarization.
License
This tool runs in full inside the Ultimate plan. The Free tier has a monthly cap.