Does MacWhisper recognise different speakers?

Yes, MacWhisper recognises different speakers, but the automatic version is part of Pro. MacWhisper's website lists speaker recognition, which it describes as telling apart and labeling different speakers in your recordings ¹. In Pro it works with local models on M-series Macs, with ElevenLabs and Deepgram, and with NVIDIA's Nemotron 3 Diarization model ². The free version lets you add up to two speakers manually ³. Our MacWhisper guide puts this in context.

This is speaker diarization: working out who spoke when, not who the speakers are. MacWhisper names them Speaker 1, Speaker 2 and so on, and you supply the real names.

Automatic labels need a compatible model

The help center says automatic recognition works with local models on the WhisperKit engine, or with the ElevenLabs and Deepgram cloud services ⁴. Compatible local models carry a green Speaker Recognition badge in the model list ⁵. WhisperKit models only run on M-series Macs, not Intel ⁶. The store page adds Nemotron 3 Diarization, an NVIDIA model that handles up to eight speakers ⁷, and Pro also lists real-time meeting transcription with speaker recognition ⁸.

It does not work in dictation, which assumes a single speaker, or in podcast mode, where each imported track is one speaker ⁹.

How to assign and fix speakers

The help center (last updated July 2025, for MacWhisper 12) describes these steps:

1. Pick a model with the Speaker Recognition badge, then transcribe.
2. Rename a speaker in the sidebar: click the name, type, press Return.
3. Merge two speakers that are really one person: right-click the name and choose Merge Speaker To.
4. Reassign a segment: hover over it and press the number key of the right speaker (1, 2, 3).

Merging matters because the app may assign parts of one person's speech to two speakers ¹⁰. Menu names may differ in newer versions.

Speaker detection not working? Check these first

It depends on the audio

Labels are the easiest to get right with two or three people, clear microphones and little crosstalk. Overlapping speech is a documented hard case for diarization in general: on the AMI meeting corpus, handling overlap explicitly cut one baseline system's diarization error rate by 20 percent relative ¹². In one test, 9to5Mac transcribed a 3-hour podcast with MacWhisper's Parakeet model and speaker recognition in 1 minute 22 seconds ¹³. That is one file on one Mac, timed by a journalist, so it says nothing about accuracy.

For whether Whisper can identify speakers on its own, see can Whisper identify speakers. Our guide to timestamps and speaker labels explains the formats.

What to do

1. Test on a real recording in the free version first, using the two-speaker manual labels, to see whether you need automatic ones. Pricing is on is MacWhisper free.
2. Upgrade only if you have three or more speakers or many files.
3. Always review labels before quoting someone. Recording calls is covered in recording meetings, and model choice in the best model.