Speaker diarization — identify and separate multiple speakers in audio.

​
Synopsis

izwi diarize <FILE> [OPTIONS]

​
Description

Analyzes audio to identify different speakers and when they spoke. Optionally includes transcription with speaker labels.

​
Arguments

ArgumentDescription
<FILE>Audio file to analyze

​
Options

OptionDescriptionDefault
-m, --model <MODEL>Diarization modelsortformer-4spk
-n, --num-speakers <N>Expected number of speakersAuto-detect
-f, --format <FORMAT>Output format: text, json, verbose_jsontext
-o, --output <PATH>Output file (default: stdout)—
--transcribeCompatibility flag (transcript output is included by default)—
--asr-model <MODEL>ASR model used for transcript generationparakeet-tdt-0.6b-v3

​
Examples

​
Basic diarization

izwi diarize meeting.wav

​
With known speaker count

izwi diarize meeting.wav --num-speakers 3

​
Transcript output (default behavior)

izwi diarize meeting.wav --num-speakers 2

​
JSON output

izwi diarize meeting.wav --format json --output diarization.json

​
Full pipeline with custom models

izwi diarize interview.wav \
  --asr-model Qwen3-ASR-1.7B-GGUF \
  --format verbose_json \
  --output interview_transcript.json

​
Output Formats

​
Text

[00:00 - 00:05] Speaker 1: Welcome to the meeting.
[00:05 - 00:12] Speaker 2: Thanks for having me.
[00:12 - 00:20] Speaker 1: Let's start with the agenda.

​
JSON

{
  "segments": [
    {"speaker": "Speaker 1", "start": 0.0, "end": 5.2},
    {"speaker": "Speaker 2", "start": 5.5, "end": 12.1}
  ],
  "speaker_count": 2
}

​
Verbose JSON (with transcription)

{
  "segments": [
    {
      "speaker": "Speaker 1",
      "start": 0.0,
      "end": 5.2,
      "text": "Welcome to the meeting."
    },
    {
      "speaker": "Speaker 2", 
      "start": 5.5,
      "end": 12.1,
      "text": "Thanks for having me."
    }
  ],
  "speaker_count": 2,
  "duration": 120.5
}

​
Available Models

ModelDescription
sortformer-4spkAlias for diar_streaming_sortformer_4spk-v2.1 (default)
diar_streaming_sortformer_4spk-v2.1Canonical Sortformer model ID

​
See Also