A complete speaker diarization pipeline using NVIDIA NeMo (via WSL2) with Whisper transcription.

• Speaker Diarization: Identify "who spoke when" in audio files
• Transcription: Optional Whisper-based transcription with speaker labels
• Dual Backend:
  • WSL2 + NeMo for advanced diarization (Linux)
  • pyannote.audio fallback for Windows
• Custom Whisper Models: Support for cached/finetuned models
• Multi-language: Support for English, Persian, Arabic, and more

from nemo_diarization import diarize_and_transcribe

result = diarize_and_transcribe(
    meeting_audio_path="meeting.wav",
    expected_language="en",
    output_transcriptions=True,
    transcriptor_model_path=None  # or path to your Whisper model
)

print(f"Speakers: {result['num_speakers']}")
print(f"Transcription: {result['transcription']}")

pip install -r requirements_windows.txt

wsl -d Ubuntu bash -c "cd /mnt/d/Git_repos/Nemo-diarization/venv_nemo_wsl && bin/pip install -r /mnt/d/Git_repos/Nemo-diarization/requirements_wsl2.txt"