Voiceprints: recognizing people across meetings

A voiceprint is a voice signature Daisy uses to recognize the same person across different meetings, so you don't have to label them every time. Voiceprints are created and stored on your own devices: your computer, and your phone if you pair one.

Enroll a voiceprint while labeling

When you name a speaker in the participants tab:

1. Click a speaker to open the Label speaker dialog.
2. Enter the person's Name (and optionally Email).
3. Leave Save voiceprint for future meetings ticked — it's on by default unless Daisy flagged this speaker as having mixed voices. This saves the person's voiceprint in your encrypted vault so Daisy can recognize them automatically.
4. Optionally tick Also check past meetings for this person to scan your older recordings for unlabeled speakers that match and label them too.
5. Click Save.

The past-meeting scan can take a few minutes if you have many recordings. If enrollment fails for any reason, Daisy still saves the name for this meeting.

Manage voiceprints

Turn on Nerd Mode in Settings → Display, then open Settings → Voiceprints. You'll see a table of enrolled people showing each person's Name, Email, number of Sessions, and number of voice Samples (more samples make matching more robust). On iPhone and iPad, manage voiceprints under Settings › Voices.

• Rename updates a person's name and email.
• Delete removes the voiceprint, so future meetings no longer auto-label that person.
• Rematch all sessions re-scans every recording for known voiceprints and back-fills speaker labels. It reports how many sessions it scanned and how many new speakers it labelled.

Enroll the same person again

If you label the same person in a later meeting and keep Save voiceprint for future meetings ticked, Daisy merges the new sample into their existing voiceprint instead of creating a second identity (it matches by name or email). More samples make matching more robust.

Long meetings and acoustic drift

A person's voice can shift across a long meeting — they move closer to or further from the mic, switch rooms, or get tired. A fresh clustering pass might split them into two groups. As long as both groups match the same vault voiceprint, Daisy collapses them back into a single identity automatically. You don't have to merge clusters by hand.

When Daisy gets a voice wrong

Two things go wrong often enough to have a control: Daisy reports "not one clear voice" on a speaker who is one person, or it splits one person into several speakers.

Open Settings → Voiceprints, expand Advanced → Speaker recognition tuning, and move the Voice matching tolerance slider:

• Daisy keeps saying not one clear voice → move it toward Looser.
• One person is being split into several speakers → move it toward Stricter.

The setting applies the next time you diarize a meeting, so re-run speaker labels on a meeting to see the effect. Sessions you already labelled keep the labels they have.

The same panel holds the Diarizer picker, which chooses the engine that groups voices. It's saved per computer, since each machine picks what its hardware can run.