Quick Start¶

Installation¶

pip install diarize

Requires Python 3.9+. All models (Silero VAD, WeSpeaker) are downloaded automatically on first use.

Basic Usage¶

from diarize import diarize

result = diarize("meeting.wav")

print(f"Found {result.num_speakers} speakers")
for seg in result.segments:
    print(f"  [{seg.start:.1f}s - {seg.end:.1f}s] {seg.speaker}")

Specifying Speaker Count¶

If you know the number of speakers, skip auto-detection:

result = diarize("meeting.wav", num_speakers=3)

Or constrain the range:

result = diarize("meeting.wav", min_speakers=2, max_speakers=5)

Exporting Results¶

RTTM format¶

# Write to file
result.to_rttm("output.rttm")

# Get as string
rttm_string = result.to_rttm()
print(rttm_string)

JSON-friendly dicts¶

for item in result.to_list():
    print(item)
# {"start": 0.5, "end": 3.2, "speaker": "SPEAKER_00"}

Pydantic serialization¶

data = result.model_dump()
# Full result as a dict, including segments, speakers, estimation_details

Iteration¶

DiarizeResult is iterable:

for seg in result:
    print(seg.speaker, seg.start, seg.end, seg.duration)

print(f"Total segments: {len(result)}")

Speaker Estimation Details¶

When auto-detection is used, diagnostic information is available:

result = diarize("meeting.wav")
if result.estimation_details:
    details = result.estimation_details
    print(f"Method: {details.method}")
    print(f"Estimated speakers: {details.best_k}")
    print(f"PCA dimensions: {details.pca_dim}")
    print(f"BIC scores: {details.k_bics}")

Logging¶

diarize uses Python's standard logging module:

import logging
logging.basicConfig(level=logging.INFO)

# Now diarize() will print progress
result = diarize("meeting.wav")

Set level=logging.DEBUG for detailed diagnostics (BIC scores per k, etc.).