Definition
VideoText editorial team. Updated September 13, 2026.
Speaker diarization answers “who spoke when?” It segments audio into speaker turns and assigns labels such as Speaker 1 and Speaker 2. It does not, by itself, name those people.
The sections below explain how speaker diarization is used in transcription and subtitle work, what people often mix up, and which VideoText workflow applies when the next step is a file or review pass.
The system finds change points in the audio, clusters segments that sound like the same voice, and writes a label per turn. A later human pass can rename Speaker 2 to “Maya.”
Interviews, meetings, and podcasts are hard to read as an unlabeled wall of text. Diarization is also a prerequisite for some caption styles that put a speaker name on each cue.
Identification matches a voice to a known person or enrollment profile. Diarization only separates voices in this recording. Teams often run diarization first, then rename labels.
Diarization returns time ranges with a speaker label attached: from 00:00:00 to 00:00:14 this was speaker A, then speaker B until 00:00:31. It says nothing about who A and B are, only that they are different voices.
Those ranges are then aligned with the recognised words to produce a transcript split into labelled turns. Errors in either step surface the same way — words attributed to the wrong person.
Similar-sounding voices, overlapping speech, and speakers moving relative to a single microphone are the main failure modes. A change in recording conditions partway through a file can cause the same person to be split into two speakers.
The opposite error is merging: two speakers with similar pitch and timbre collapsing into one label. Both are normal corrections during review, and neither means the words were wrong.