Donna
Features How it works Roadmap Pricing FAQ
Download for Mac

Guides / Speaker Identification

What is speaker diarization in meeting transcripts?

Updated 2026-11-09

Speaker diarization is the process of working out who spoke when in an audio recording. A model splits the audio into stretches of speech, measures the voice in each stretch, and groups similar voices together, so a transcript can show Speaker 1, Speaker 2 and so on instead of one unbroken block of text. It labels voices, but it does not know names unless something else supplies them.

How speaker diarization works

Diarization answers one question: who spoke when. It does not decide what was said. That is the job of speech recognition, and the two outputs are lined up by timestamp to give you a transcript with a label on every line.

Most systems follow the same broad path:

  • Find the speech. The audio is scanned for stretches where someone is talking, skipping silence and background noise.
  • Cut it into segments. Those stretches are split wherever the voice seems to change.
  • Fingerprint each segment. A model turns each segment into a compact numeric description of the voice, often called an embedding.
  • Group the fingerprints. Segments that sound alike are clustered together, one cluster per person.
  • Attach labels. Each cluster becomes Speaker 1, Speaker 2 and so on, and the transcript lines are tagged to match.

The system usually does not know in advance how many people are on the call, so it has to estimate the number of voices from the audio alone.

Diarization is not the same as knowing who someone is

Plain diarization gives you anonymous labels. It can tell you that the voice at minute 4 is the same voice at minute 31, but not that it belongs to Priya. Turning Speaker 2 into a name is a second step, usually called speaker identification. Either you name the voice yourself, or the tool matches it to a person it already knows.

That second step is what makes a transcript usable. Action items need owners, and a note saying Speaker 2 will send the contract is not much help a week later. For the practical side, see how to label who said what in a meeting transcript.

Where diarization goes wrong

Even good systems stumble in predictable places:

  • Overlapping speech. When two people talk at once, a line can be credited to the wrong person or to only one of them.
  • Similar voices. Two people with close vocal ranges can be merged into one speaker, or one person can be split into two.
  • Short interjections. A quick yes or right gives the model very little voice to work with.
  • Poor audio. Echo, a laptop mic across a room, or several people sharing one microphone all blur the fingerprints.

A few habits help. Use a decent mic, give each remote person their own audio line where you can, and skim the labels right after the call while you still remember who was there.

How Donna does it on your Mac

Speaker Identification in Donna runs on-device. She tells the voices in a recording apart and matches them to the people on the calendar invite.

The rest of the pipeline stays local too. Recording, Whisper transcription and speaker matching all happen on your Mac. No bot joins the call, nothing appears in the participant list, and nothing is uploaded. The details are in the privacy policy.

Donna captures your mic and the call audio, and it works on Zoom, Google Meet, Microsoft Teams, Slack huddles, FaceTime, Discord and Webex. You can also click any transcript line to hear that moment, which is a quick way to confirm who said something before you quote them.

Fixing a name teaches Donna

If a voice is labeled wrong or left unnamed, you fix it once and Donna carries it forward:

  • Name the voice once and she recognizes it in every future meeting.
  • She also relabels your past transcripts with that name.
  • Any correction you make teaches her.
  • Renaming a speaker makes her rewrite the meeting summary, so decisions and action items credit the right people.

This pays off most on recurring calls like standups and 1:1s. Name your teammates once and the transcripts stay readable. Speaker Identification is part of Pro, at $8 a month or $80 a year. See pricing for what Free and Pro include.

When to trust the labels and when to check them

Diarization works best on small calls where people have distinct voices and take turns. It gets harder on big panels, crosstalk-heavy discussions, and a conference room where everyone shares one microphone.

Whatever tool you use, check the labels before you quote someone or send action items to the wrong person. A thirty-second skim right after the call is usually enough, and each correction makes the next transcript better.

Try Donna. Records, transcribes and summarizes your meetings on your Mac, with nothing sent to a server.

Download Donna for Mac

Frequently asked questions

Is speaker diarization the same as speaker recognition?

Not quite. Diarization separates voices and gives them anonymous labels, while speaker recognition or identification ties a voice to a known person. Donna does both on your Mac: she tells voices apart, then matches them to the invite and to names you have given her.

Does speaker diarization need the cloud?

Many services send audio to a server for it, but it does not have to work that way. In Donna it runs on your Mac, and audio and transcripts never leave it.

What if a line is credited to the wrong person?

Fix the name. In Donna, naming a voice teaches her, relabels past transcripts, and a speaker rename also rewrites the summary so credit lands on the right person.

Do I need Pro for Speaker Identification in Donna?

Yes. Speaker Identification comes with Pro, along with recording, transcripts and summaries, at $8 a month or $80 a year. The Free plan covers the calendar features and the on-device chat.

Can speakers be told apart without a bot joining my call?

Yes. Donna records on your Mac, so no bot joins and nobody sees a recorder in the participant list. She does remind you to get consent where it is required.

Related guides

  • How do you label who said what in a meeting transcript?
  • How do I fix a wrong speaker name in a transcript?
  • What is the keyboard shortcut to copy a Google Meet link?
  • How do I open Microsoft Teams meetings automatically on a Mac?
Donna

The Mac assistant that opens your meetings, records and transcribes every call on-device, and keeps your whole day on track.

Product

  • Features
  • How it works
  • Roadmap
  • Pricing

Get Donna

  • Download for Mac
  • Integrations
  • FAQ

Company

  • Privacy
  • Terms
  • Contact
© 2026 Donna. Made for people with too many meetings.Designed in the menu bar.