What is speaker diarization?
Quick answer
Speaker diarization is the part of a transcription tool that listens to a recording and works out who spoke when, splitting the audio into turns by voice. It answers "how many voices, and which one is talking now", and the result is the speaker tags you see on each line.
Last updated
Take the free module →Module 31 of the free AI Apps Worth Knowing course, about 10 minutes. No account needed.
Updated · By Robert Breen
Why it matters for a small business
A transcript with no speakers is hard to use for client work. If an interview says "the fee was a surprise", you need to know whether the client said it or your own staff did, because one is a complaint and the other is a guess. Fixing that by hand means replaying the audio and marking every turn, which is slow work nobody budgets for.
Diarization does that marking automatically. It does not know names, though. It hears that two voices differ and calls them Speaker 0 and Speaker 1. Similar voices, crosstalk and a poor microphone can merge two people into one or split one person into two. That is why you check the speakers against what you know before you quote anyone.
In a real lesson: Speak AI: Turn Customer Interviews into Themes and Quotes
In the free Speak AI lesson, the Harbor Lane Accounting interview is two computer voices reading a made-up script. When the transcript opens, Speak AI has split the call by voice: Speaker 0 is Dana, the owner, and Speaker 1 is Mike, the client. Each turn has a play button and a time, such as 00:25.
The Insights tab adds that this is A 1-on-1 conversation, and the panel has a separate Speakers tab. When you export to Word, Display speaker names is on, so the file keeps who said what. Knowing Speaker 1 is the client is what lets AI Chat quote Mike's complaints instead of Dana's questions.
Try this lesson free or read the step-by-step guide.
Common confusions
Speaker diarization vs speaker labels
Diarization is the process that splits a recording by voice. Speaker labels are its output on the page, such as "Speaker 1:" or a name someone added.
Speaker diarization vs speech-to-text
Speech-to-text writes down the words. Diarization decides whose words they are. Most tools run both on the same upload.
Tips
- Write down which speaker number is which person as soon as you open a transcript.
- Use a separate microphone per person on important calls; clearer voices split better.
- Listen to a turn or two where people interrupt each other, since that is where splits go wrong.
Related terms
More Voice agents terms
Where you use it: free lessons
- Speak AI: Turn Customer Interviews into Themes and Quotes (Speak AI, 10 min)
- Summarize a Meeting Transcript with ChatGPT (ChatGPT, 9 min)
AI tools that use it
Frequently asked questions
- Does diarization know people's names?
- No. It tells voices apart and numbers them. Names come from you, from a meeting tool that knows who joined, or from a later step.
- Why does my transcript show one person as two speakers?
- A change of microphone, background noise or a long pause can make one voice sound like two. Check those spots against the audio.
- Is speaker diarization the same as voice recognition?
- Not quite. Diarization only separates voices within one recording. Voice recognition tries to match a voice to a known person.