Speaker identification

After a recording finishes, Perchnote automatically figures out who spoke when. This is neural diarization (pyannote-grade) running on CoreML, and it splits turns at the word boundary, so fast back-and-forth keeps each speaker on their own line instead of mashing two voices into one. It groups the audio into speakers and labels them Speaker 1, Speaker 2, and so on.

To give speakers real names, open the transcript drawer (Cmd+T) and rename them. Once you name a speaker, Perchnote remembers the voice and can match it in future recordings.

Settings (Settings, then Audio):

• Auto-diarize. On by default. Runs speaker detection automatically after each recording.
• Re-detect. If you turn auto-diarize off, or the results need a redo, there is a button to run speaker detection again on demand.

Stereo recording (you on one channel, everyone else on the other) improves separation, so turn it on if speaker accuracy matters to you.