Repository navigation
This marks the initial stable release of the Speaker Diarization & Splitting System!
This powerful AI-powered Python script is designed to automatically identify individual speakers in an audio file and split their speech into separate, clean tracks.
Key Features of this Release:
- 🎙️ Multi-Speaker Diarization: Accurately identifies and separates an unlimited number of speakers.
- 🧹 Crosstalk Removal: Generates pristine, non-overlapping audio tracks for each speaker, ensuring only one voice per track.
- ⚙️ Batch Processing: Automatically handles multiple audio files (
.wav,.mp3,.m4a,.flac) placed in theaudio/pendingdirectory. - 🚀 GPU Acceleration (NVIDIA CUDA): Automatically detects and leverages NVIDIA GPUs for significantly faster processing.
- 🗣️ Flexible Speaker Count: Supports automatic detection, exact speaker counts, or min/max ranges.
- 🤫 Verbose/Quiet Mode: Choose between detailed logs for debugging or clean output for general use.
Automated Sanity Checks for a Smooth First Run:
This release includes built-in checks to guide users through the setup process:
- FFmpeg Detection: Provides instructions and opens the download page if FFmpeg is missing.
- Hugging Face Model Access: Proactively checks for accepted user agreements for
pyannotemodels, providing direct links if acceptance is pending.
Installation & Usage:
We highly recommend using uv for installation for its speed and reliability. Please refer to the README.md for detailed installation instructions, including specific commands for GPU and CPU users, and how to set up your Hugging Face token.
Project Links:
- Homepage: https://github.com/lukium/speaker-diarization-system
- PyPI: https://pypi.org/project/speaker-diarization-system/
- Bug Tracker: https://github.com/lukium/speaker-diarization-system/issues
Thank you for checking out this project!
Full Changelog: https://github.com/Lukium/speaker-diarization-system/commits/1.0.2