How prince1314-patel/multi-speaker-diarization works
Project type and purpose
• This is a Python‐based, modular multi-speaker meeting transcription system.
• It ingests audio in various formats, performs speaker diarization and ASR transcription using NVIDIA NeMo, WhisperX, and pyannote-audio, then exports results. An optional Streamlit UI provides upload and visualization.
File structure
• Top-level:
• .gitignore
• README.md
• Implicit subcomponents (in code):
• Ingestion module (handles .wav, .mp3 via librosa, pydub, ffmpeg)
• Diarization engines (pyannote-audio, NeMo, WhisperX)
• Transcription engine (ASR via nemo_toolkit[asr])
• Configuration (omegaconf, python-dotenv)
• Output exporters (.txt, .csv, .json, .srt/.vtt)
• Optional Streamlit frontend for UI
• Tests (pytest)
README for additional insights
• Describes modular, secure, scalable design.
• Two branch strategies: pure pyannote-audio vs hybrid whisperX+pyannote.
• Key dependencies define technology choices.
Elements
• Clear labels: e.g., “Audio Ingestion”, “Speaker Diarization”, “ASR Transcription”
• Blue rectangles for core processing modules
• Orange parallelograms for input/output components