# Speaker Diarization

> Working out who spoke when in a recording. Each line gets a speaker tag.

**Field:** Artificial Intelligence  
**Category:** Multimodal  
**Also known as:** Diarization, Speaker Labeling, Who Spoke When

## Think of it like

It is like a play script. Each line has the name of who says it.

## Definition

Speaker diarization works out who spoke when in a recording. It splits the audio into stretches and gives each stretch a speaker label, such as Speaker 1 and Speaker 2. It does not know people's names unless it is told. It is often paired with speech-to-text, so a transcript shows who said each line. People talking over each other is the hardest part.

## Example

A team records a one-hour call with four people. The transcript comes back with each line marked Speaker 1 to Speaker 4, and Jack renames them.

## Example code

```text
[00:00:04] Speaker 1: Shall we start with the budget?
[00:00:07] Speaker 2: Yes, I have the numbers here.
```

A transcript after diarization. Each line has a time and a speaker label.

## Why it matters

A transcript without speakers is hard to follow. Meeting notes, interviews and court records all need to know who said what.

## Common mistake

Expecting real names. The tool tells voices apart, but a person or another step has to say who each voice is.

## Three principles

- **Who spoke when.** Each stretch gets a label.
- **Labels, not names.** Speaker 1 until someone says.
- **Overlap is hard.** Two voices at once confuse it.

## Related terms

- [Speech-to-Text (Speech Recognition)](https://learn.benibix.com/term/ai/speech-to-text)
- [Voice Activity Detection (VAD)](https://learn.benibix.com/term/ai/voice-activity-detection)
- [Voice Agent](https://learn.benibix.com/term/ai/voice-agent)
- [Named Entity Recognition (NER)](https://learn.benibix.com/term/ai/named-entity-recognition)

## Told apart from

- [Speaker Diarization vs Speech-to-Text](https://learn.benibix.com/compare/speaker-diarization-vs-speech-to-text): People think speech-to-text already says who spoke. It does not. Speech-to-text writes down the words. Speaker diarization splits the sound by voice and labels each part Speaker 1, Speaker 2 and so on. A meeting script needs both.

![Top row: a voice goes through three boxes, speech to text, a language model, and text to speech, then out as a voice. Bottom row: a voice goes through one box and straight out as a voice. Under the top row, small tags mark the wake word, voice detection and who is speaking.](https://learn.benibix.com/diagrams/ai5-speech-pipeline.svg)

---

Source: https://learn.benibix.com/term/ai/speaker-diarization
Author: Behnia Zarei (a developer in Vienna, Austria, who builds and teaches data, AI and business intelligence at BENIBIX)
Publisher: BENIBIX Learn (https://learn.benibix.com/)
Last updated: 2026-11-17
Licence: CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). Quote freely with attribution.
