Glossary: Meetings

What is Diarization?

Working out who spoke when in a recording, so the transcript can label each line with a speaker.

Updated Nov 18, 2026By Bruce W.

Speaker diarization is the step that splits a recording by voice and answers “who spoke when?” Speech-to-text writes the words. Diarization decides whose words they are. The result is speaker labels.

How it works, roughly

  1. The audio is cut into short stretches.
  2. Each stretch gets a “voiceprint”, a numeric fingerprint of how the voice sounds.
  3. Similar voiceprints are grouped, and each group becomes Speaker 1, Speaker 2 and so on.
  4. Names come later, from a person telling the software, a calendar invite, or a voice it already knows.

Diarization vs speaker identification

The two are often mixed up, but they answer different questions:

  • Diarization asks “is this the same voice as before?” It groups the recording by voice and labels the groups Speaker 1, Speaker 2 and so on, without knowing who anyone is.
  • Speaker identification asks “whose voice is this?” It matches a voice to a person it already knows, so the label can say “Priya” instead of “Speaker 2”.

Most meeting tools do diarization first and then put names on the groups, either from a voice they have heard before or from you telling them once. That’s why fixing one wrong name can correct a whole transcript. We walk through it in How to label speakers in a meeting transcript.

What makes it hard

Crosstalk, similar voices, and one microphone shared by a room full of people.

An example

Without diarization, a transcript is one block: “We should move the launch. I agree, the docs aren’t ready. Can we do the 21st? The 21st works.”

With it, the same words become a conversation:

  • Speaker 1: We should move the launch.
  • Speaker 2: I agree, the docs aren’t ready.
  • Speaker 1: Can we do the 21st?
  • Speaker 3: The 21st works.

Now you can see who pushed for the change and who agreed, which matters for action items and follow-ups.

Tips for better results

  • Use a headset or your computer’s own audio for online calls, rather than a laptop mic in a busy room.
  • Avoid talking over each other when a decision is being made.
  • Fix a wrong name once. Good tools remember the correction.
  • In person, put the laptop in the middle of the table.

Diarization is judged separately from word error rate. A transcript can get every word right and still give a line to the wrong person.

In Fetch

Fetch records your microphone and your computer’s sound separately, so your lines are always yours. It also learns your contacts’ voices on your own computer, so the transcript names them next time. See Speaker labels.

Questions people ask

Is diarization the same as speaker labels?

Close. Diarization is the process. Speaker labels are the result you see, like 'Priya:' or 'Speaker 2:'.

Why does my transcript mix up two speakers?

Usually because the voices sound alike, people talked over each other, or everyone shared one microphone. Correcting a wrong label once helps tools that learn voices, and a separate microphone for each person helps most.

Still scratching your head?

Send us a note and a real person will get back to you.

Contact support