Speaker identification splits a transcript by voice, so a room full of people reads as a conversation instead of one long block of text.
It matters most for meetings in a room. On a video call, each person usually arrives on their own audio channel. Around a table, everyone is one microphone, and without this feature you cannot tell who said what.
This is a Pro feature. Recording an in-person meeting works on the free plan. Putting names to the voices is what Pro adds.
It runs on your Mac
Like transcription, this happens on your own machine. The model is compiled into the app, so there is nothing extra to download and nothing to send anywhere. Pro unlocks a local capability; it does not move your meeting to a server.
It works after the meeting, not during
Speaker identification is batch only. It runs once the recording stops. You will not see speaker labels appearing live during a call.
Turning it on
Settings, AI, Transcription. Choose Local Diarization.
If you are on the free plan, the option is visible but locked, and selecting it takes you to the upgrade screen.
(“Diarization” is the technical word for splitting a recording by speaker. It means the same thing as speaker identification.)
Naming people
After a meeting is transcribed, open the transcript. Each block is headed with a speaker label: Speaker 1, Speaker 2 and so on.
- Select the speaker label above any block.
- Type the person’s name, or pick someone you have named before.
- Choose whether to apply it to just this block or to every block from that speaker in the meeting.
Almost always you want every block. Use the single-block option when the split got one passage wrong and you are correcting just that.
What your coding agent sees
get_transcript returns the transcript with speakers attributed. Real names appear once
you have set them and are on Pro. Otherwise your agent sees the generic labels.
See the MCP tool reference.
Free plan
Everything else about the recording works: the audio is captured, the transcript is written, the summary is written. The transcript is simply not split by person.