Skip to main content
Speaker identification (diarization) splits a transcript by voice and labels each line with a speaker. You can then name the speakers, and summaries can say who decided what. It works in meeting notes, the Transcribe page and watch folders. Speaker identification is a Pro feature.

Turn it on

The setting is Identify Speakers, “Label different speakers in the transcript”. It appears in:
  • Notes › Notes settings › Recording Defaults, for every meeting you record.
  • Transcribe › Transcribe settings › Defaults, and in the Transcribe dialog for each file.
  • The inspector’s Details tab on a note or transcription, to re-run an existing recording.
  • Each watch folder.
It only appears when the selected transcription model can identify speakers. Turn it on before you record, or later and click Regenerate Transcript. Notes settings, Recording Defaults tab, with Identify Speakers, Number of Speakers and Speaker model

Options

Automatic uses Nemotron-3. It falls back to Standard when you pin a speaker count, when a meeting has more than 8 people, or while the Nemotron-3 model is still downloading.

Supported models

Identified on your computer

For these models Vowen identifies speakers itself, on your Mac or PC. No audio leaves your computer for this step. On macOS the speaker models download in the background the first time you use them. On Windows, the first diarization downloads a model bundle of about 30 MB. On Windows, meeting notes identify speakers per audio track: the other side of the call from the system-audio track, the people in your room from the microphone track. If the tracks are missing, Vowen uses the combined recording.

Identified by the cloud provider

With Identify Speakers on, OpenAI transcription uses gpt-4o-transcribe-diarize instead of gpt-transcribe. That model accepts no vocabulary hints and no language setting, so both are dropped for that transcription.
Gemini has no diarization setting. Vowen asks for speaker labels in the prompt. If Gemini returns text without labels, the transcript is saved without speakers.
Sarvam, Cartesia, OpenRouter, Nemotron and custom speech servers cannot identify speakers. Nemotron fails with “Speaker diarization is not supported by Nemotron.”

How speakers are labeled

  • Lines split where the speaker changes, word by word.
  • Very short or silent clips come back as one speaker.
  • If you have connected Google Calendar, the meeting’s attendees who did not decline set the maximum number of speakers.
  • If identification fails, the transcript is kept and a warning above it reads “Speakers could not be identified: ”.

Name the speakers

Speakers start as Speaker 1, Speaker 2 and so on.
1

Open the Speakers tab

Open the note or transcription and click Speakers in the inspector. It appears when the transcript has speaker labels.
2

Type a name

Vowen suggests names you have used before in notes and transcriptions. Use the arrow keys and Enter to pick one. Tab moves to the next speaker.
3

Save

Click Save Changes, or press Cmd + Enter / Ctrl + Enter. Cancel discards your edits. When nothing is pending the tab shows Saved.
The Speakers tab in the inspector with named speakers Named speakers appear in the transcript, in summaries you regenerate, and in the People filter on the Notes page. To rename a person across every note at once, use the People filter. With voice profiles on, Vowen can name people it has heard before.

Merge two speakers

Sometimes one person is split into two labels.
1

Open the Speakers tab

Merging is available when there are at least two speakers and the note is not recording.
2

Open the menu on the speaker to remove

Click the … button, “Merge into another speaker”.
3

Pick the speaker to keep

Under Merge into, choose the target.
4

Confirm

Click Merge in “Merge into ?”.
A speaker merge cannot be undone. Every line moves to the target speaker. If the target has no name, it takes the source’s name.
To move one line, use Reassign this line to in that line’s Segment actions menu. See Summaries.

Tips

  • Use a good microphone. Overlapping speech is hard for every model.
  • For short clips, pin the Number of Speakers.
  • Try the on-device models first. They cost nothing per minute and keep audio on your computer.