A comprehensive audio analysis toolkit with both command-line scripts and a web interface for audio feature extraction and music classification. The application detects instruments and mood/theme tags in music files.
- Instrument Detection: Identifies up to 40 different instruments present in the audio
- Mood/Theme Classification: Classifies tracks into mood categories (happy, sad, relaxed, aggressive)
- Audio Metrics Analysis: Extracts danceability, BPM, and musical key
- Command-line Scripts: Run analysis directly from the terminal
- Web Interface: Easy-to-use browser-based interface for uploading and analyzing audio files
- Batch Processing: Analyze multiple audio files in batches
- CSV Export: Download analysis results in CSV format for further processing
music-feature-extraction/
├── models/ # Pre-trained models
│ ├── mood_model/ # Mood detection models
│ └── instrument_model/ # Instrument detection models
├── scripts/ # Analysis scripts
│ ├── classify_instruments.py
│ └── classify_mood_theme.py
├── web/ # Web interface
│ ├── src/ # JavaScript sources
│ │ ├── main.js # Core application logic
│ │ ├── audioUtils.js # Audio processing utilities
│ │ ├── featureExtraction.js # Feature extraction worker
│ │ ├── inference.js # Model inference worker
│ │ └── viz.js # Visualization components
│ ├── models/ # Client-side models
│ ├── index.html # Main HTML page
│ └── style.css # CSS styles
├── data/ # Directory for audio files to analyze
└── results/ # Analysis results are saved here
- Python 3.12: For command-line scripts
- TensorFlow: For the pre-trained models
- Essentia: Audio processing and feature extraction
- Modern Web Browser: Chrome, Firefox, or Edge (for web interface)
-
Clone the repository:
git clone https://github.com/jasonmokk/music-feature-extraction.git cd music-feature-extraction -
Create a Python virtual environment and install dependencies:
python3.12 -m venv venv source venv/bin/activate # On Windows, use: venv\Scripts\activate pip install -r requirements.txt
-
Install Node.js dependencies (if you want to run the web interface locally):
npm install npm start -
Access the web interface at:
http://localhost:3000
-
Place your audio files (MP3, WAV, or OGG format) in the
data/directory. -
Run the mood/theme classification script:
python scripts/classify_mood_theme.py -
Run the instrument detection script:
python scripts/classify_instruments.py -
Find results in the
results/directory:instrument_predictions.csv: Contains instrument detection resultsmood_theme_predictions.csv: Contains mood/theme classification results
-
Upload your audio files using the file uploader or by dragging and dropping them onto the upload area.
-
For batch uploads, click on the upload area and select "Upload Folder" to process multiple files at once.
-
The analysis will automatically start and show progress as files are processed.
-
View the results for each file, including:
- Mood analysis (happy, sad, relaxed, aggressive)
- Danceability metric
- BPM (Beats Per Minute)
- Musical key
-
Download all results as a CSV file using the "Download Results as CSV" button.
The web interface supports processing files in batches:
- Upload up to 1000+ files at once
- Files are processed in batches of 20 to optimize memory usage
- Progress bar indicates overall completion percentage
- Results are combined at the end for a single CSV export
To use your own custom models:
- Place TensorFlow models in the appropriate directories under
/models - Update the model configuration in
src/main.js(for web interface) or in the Python scripts
- For large audio files, the system automatically shortens audio to 15% of its original length
- Worker threads handle parallel processing to maximize performance
- Memory is automatically freed between batches
- Audio decoding errors: Ensure your audio files are valid MP3, WAV, or OGG format
- Memory issues with large batches: Try reducing the batch size in
web/src/main.js - Model loading errors: Check that all required model files are present in the models directory
The web interface requires a modern browser with support for:
- Web Audio API
- Web Workers
- ES6+ JavaScript features
This project is licensed under the MIT License - see the LICENSE file for details.
- Essentia team for their audio analysis tools
- TensorFlow.js for browser-based machine learning
- WaveSurfer.js for audio visualization