Skip to content
← Help

Identifying who said what

Daisy can group a meeting's voices into separate speakers and let you put names to them. The grouping (diarization) runs entirely on your machine, alongside transcription — no cloud account required.

Group the speakers

If a recording has no speaker labels yet, the Speakers row offers to create them.

  1. Open the meeting and select the transcript tab.
  2. In the Speakers row, click Diarize speakers.

Daisy clusters the other participants' voices locally and labels them automatically as Person A, Person B, and so on. Once labeled, each speaker appears as a chip in the Speakers row. A chip you have named yourself is shown with a stronger outline so you can tell named speakers from auto-labeled ones.

Name a speaker

  1. Click a speaker chip to open the Label speaker dialog.
  2. Click Play sample audio to hear a roughly five-second clip of that voice so you can identify it.
  3. Enter a Name, and optionally an Email.
  4. Click Save.

To remove a name and return the speaker to its automatic label, open the dialog and click Clear label.

Re-diarize

If the grouping looks off, click Re-diarize in the Speakers row. This re-clusters the voices on your machine from scratch and resets the automatic labels. Your manual names are kept and re-applied where Daisy can match them. A live progress bar shows how far along the re-run is.

The Participants tab

For a structured view, open the Participants tab on the meeting. Daisy splits the speakers into:

  • Voice Matched — auto-paired to a voiceprint in your vault. Labels carry over to future meetings.
  • Manually Added — names you typed in for this session. No enrolled voice sample yet.
  • From Invite — calendar attendees who didn't speak during the meeting (or whose voice didn't match anything). One click adds them as a manual participant.

Add a participant the diarizer missed

If Daisy didn't pick someone up, use + Add participant in the Participants tab. You can either:

  • Pick a voice from your vault gallery — useful when you have a voiceprint for them but they were too quiet for the diarizer to cluster.
  • Type a name directly — quick when no voiceprint exists yet.

Participants you add this way live only on the current session — removing one doesn't touch the vault or any other meeting.

Good to know

  • Grouping is not perfect. Review the results and correct any chips that were merged or split incorrectly.
  • Your manual names are kept across re-diarizes and applied to future meetings via voiceprints.
  • Long meetings where the same person's voice drifts across the call collapse back to one identity automatically when both clusters match the same vault voiceprint — see Voiceprints.