Speaker diarization

Speaker diarization splits a transcript by voice, so each line shows who said it. It runs on your Mac, after transcription. Diarization is a Pro feature. For background on how it works, see speaker diarization in TranscribeX.

Before you start

• The job must be finished. Diarization only runs on a completed transcription.
• The model must give timestamps. Speakers are matched to lines by time. Some models (marked “Timestamps not supported” in their model list) produce text only. For those jobs the app shows “…doesn’t support timestamps, so speaker diarisation can’t be applied to this transcript.” Pick a model with timestamps, such as Whisper or Parakeet. See choosing a model.

Diarize a finished transcript

1. Open the job from your library.
2. In the right panel, choose the Overview tab, then the Speakers segment.
3. Turn on Show Speakers.
4. Click Start Diarisating. A progress bar shows while it runs.

When it finishes, each speaker appears in the list as Speaker 1, Speaker 2 and so on, with the number of lines they have in brackets. If a voice matches one of your saved voice profiles, that person’s name and color are used instead. “Diarisation uses all saved voice profiles automatically.”

Note: Running diarization again removes all current speakers and their assignments, including names you typed. Rename speakers after your final run.

Manage speakers in the Speakers panel

Each row in the speaker list has these controls:

• Checkbox: selects the speaker. Selected speakers are shown in the transcript and are the ones used by Merge Speakers and Save as Voice Profiles. Select All toggles every speaker at once.
• Name field: type to rename the speaker. The new name appears on every line.
• Person menu: appears when you have saved speakers. Pick a name from Manual Speakers or Voice Profiles to apply that name and color in one click.
• Color well: change the speaker’s color.
• Play button: plays a short sample of the voice, so you can tell who it is.
• Minus button: removes the speaker. Their lines move to the first speaker in the list.

Below the list:

• Add Speaker adds an empty speaker you can assign lines to.
• Merge Speakers combines two or more selected speakers. Choose which speaker to keep, then click Merge. The others are merged into it and deleted.
• Save as Voice Profiles saves the selected speakers so future jobs recognize them. See voice profiles.

Fix who said a line

Diarization is rarely perfect. To change the speaker of one paragraph, click the speaker name above it, or the person icon in its hover toolbar (Assign speaker), and pick the right person.

To reassign many lines at once:

1. Hover a paragraph and click the checkbox icon (Select). The transcript enters selection mode.
2. Tick the other paragraphs you want, or use the checkbox at the top to select all.
3. Click Assign Speaker and choose the speaker.
4. Click Exit when you’re done.

Diarize meetings automatically

For microphone and app recordings, diarization can run by itself after transcription. Turn on Settings › Automation › Auto Speaker Diarisation after transcription (Pro). It’s on by default for Pro users. See meeting automation.

Diarize many files at once

When you create a bulk transcription (Pro), turn on Enable speaker diarisation in the Speaker Diarisation section. Every file in the bulk is diarized with your saved voice profiles. See bulk transcription.

Live diarization

Live transcription can also show speaker changes and recognize saved voice profiles while you record. Turn on Settings › Live › Diarize Live Transcription (Pro), or Enable Diarization in the live Caption Settings. It works with supported live models. See live transcription.

Cloud models

Mistral and ElevenLabs cloud models have their own Diarize switch in their settings pane. When it’s on, the provider labels speakers and TranscribeX skips its own diarization. Speakers from Mistral can’t be saved as voice profiles, because Mistral doesn’t return voice data. With Mistral diarization on, reflowing subtitles also won’t work, because Mistral doesn’t provide word timestamps.