Get a speaker activity timeline

POST /v1/speaker/diarizations

Returns who spoke when, without transcription. Speaker endpoints accept uploaded bytes rather than arbitrary local paths, and num_speakers is clustering guidance, not an identity claim.

Guide and examplesTranscription and speakers·All endpoints

Request body

multipart/form-dataRequired

Responses

{
  "duration": 10,
  "speakers": [
    "SPEAKER_00",
    "SPEAKER_01"
  ],
  "segments": [
    {
      "start": 0.4,
      "end": 3.2,
      "speaker": "SPEAKER_00"
    },
    {
      "start": 4,
      "end": 8.1,
      "speaker": "SPEAKER_01"
    }
  ]
}