Speaker diarization
Speaker diarization is the task of splitting an audio recording by speaker, labelling each stretch of speech with who said it. It answers the question "who spoke when".
Also called: diarisation, diarization, speaker labelling, who spoke when
By Divya Sharma · Checked Nov 15, 2026 · How we review
A plain transcript of a meeting is a wall of words. Diarization adds labels such as Speaker 1 and Speaker 2 to each segment, so you can follow the conversation, attribute quotes and summarise each person's points. It is a standard feature of meeting notes, call-centre analytics, interview transcription and podcast tools.
How it works
A typical pipeline first finds where speech occurs, then cuts it into short segments, turns each segment into a speaker embedding (a compact voiceprint) and clusters the embeddings so that segments with similar voices share a label. Newer neural systems handle more of this in one model, including overlapping speech. Open toolkits such as pyannote, with models like pyannote community-1, are widely used; some transcription models output speaker labels directly.
How it differs from speech-to-text
Speech-to-text produces the words. Diarization produces the speaker turns, without knowing what was said. Products merge the two by aligning timestamps. Diarization also does not identify people by name: it groups voices. Matching a voice to a known person is speaker identification, a separate task with its own privacy rules.
Limits
Errors rise with crosstalk, short interjections, similar voices, poor microphones and an unknown number of speakers. Many tools let you give the expected speaker count, which helps.
Related terms
- Speech-to-text
- Speech-to-text, also called automatic speech recognition (ASR), is software that converts spoken audio into written text. It powers transcription, captions, voice commands and the listening half of voice agents.
- Voice agent
- A voice agent is an AI agent that holds a spoken conversation in real time, listening to a person, deciding what to say or do, and answering in a synthetic voice, often over a phone line or in an app.
- Embedding
- An embedding is a list of numbers (a vector) that represents a piece of text, an image or other data, produced so that items with similar meaning end up with vectors close together.
- Text-to-speech
- Text-to-speech (TTS) is software that turns written text into spoken audio. Modern TTS uses neural networks that predict the sound of speech from the text, then render it as a waveform.