Speaker diarization

Separate a recording into speaker turns and align speaker labels with the transcript, either on its own or as one stage of a pipeline.

Speaker diarization answers who spoke when. It segments audio into speaker turns and associates those turns with anonymous labels such as speaker_0 and speaker_1.

Diarization is not identification

Diarization separates voices within a recording. It does not, by itself, know that a voice belongs to a specific person. Use Speaker identification when your application needs a known identity.

Where it helps

  • Contact-center and sales-call analytics
  • Meetings and interviews
  • Podcasts and media archives
  • Clinical or research conversations
  • Any workflow where speaker attribution changes meaning

Evaluate difficult turns

Pay special attention to overlapping speech, backchannels, very short turns, similar voices, long silences, and recordings that mix channels. A useful evaluation scores both segmentation and transcript attribution.

Combine with ASR

Run diarization in the same pipeline as transcription so timestamps and speaker turns can be aligned before formatting or downstream understanding runs.

On this page