An evolving audio processing pipeline designed to separate and enhance audio components using open-source tools. The project aims to provide a modular framework for working with raw audio files, enabling separation, refinement, and post-processing.
🚀 Current Features
- Audio Input Handling: Uses librosa for reading and handling audio files.
- Preprocessing: Includes resampling, normalization, and trimming using librosa.
- Audio Classification: Utilizes panns-interference model trained on Google's Audioset to classify audio content.
- Source Separation: Implements Demucs for music source separation.
- Noise Reduction: Enhances audio by removing background noise using DeepFilterNet.
- Post-Processing: Uses librosa to save processed audio files.
- Modular Architecture: Designed for easy extension and customization.
📁 Project Structure
freqsplit/
├── api/ # API implementation (django, redis, celery)
├── client/ # Client-side interactions (vite, react, typescript)
├── LICENSE # License file
├── pyproject.toml # Project metadata and dependencies
├── pytest.ini # Pytest configuration
├── README.md # Project documentation
├── requirements.txt # Python dependencies
├── src/ # Core processing modules
│ ├── freqsplit/ # Main package directory
│ │ ├── __init__.py # Package initialization
│ │ ├── input/ # Audio input handling
│ │ ├── preprocessing/ # Normalization, resampling, trimming
│ │ ├── separation/ # Source separation with Demucs
│ │ ├── postprocessing/ # Post-processing and saving results
│ │ ├── refinement/ # Spectrogram-based enhancement
│ │ ├── spectrogram/ # Spectrogram generation and analysis
├── tests/ # Unit and integration tests
└── venv/ # Virtual environment (should be excluded from version control)For detailed instructions on installing dependencies, setting up Python environments, and configuring the project, visit the Wiki.
This project is licensed under the Apache License 2.0.