Sound Stamp

A deep learning project for music tagging to categorize tracks by genre, mood, and more.

The model is based on the 5-layer fully convolutional neural network (FCN) from Choi et al. (2016) Automatic tagging using deep convolutional neural networks. It is trained on the top-50 tags from the MagnaTagATune data set, which contains 25.863 tracks.

Installation

1. Clone the repository:

git clone https://github.com/hoverslam/sound-stamp

2. Navigate to the directory:

cd sound-stamp

3. Set up a virtual environment:

# Create a virtual environment
python -3.11 -m venv .venv

# Activate the virtual environment (Windows)
.venv\Scripts\activate

# Activate the virtual environment (Unix or MacOS)
source .venv/bin/activate

4. Install the dependencies:

pip install -r requirements.txt

Usage

Train a music tagger:

python -m scripts.train -f "music_tagger"

-m , --model        Name of the file to save the model. Defaults to 'music_tagger'.

Evaluate the performance on Area under the ROC Curve:

python -m scripts.evaluate -f "music_tagger"

-m , --model        File name of the trained model. Defaults to 'music_tagger'.

Tag an audio track:

python -m scripts.tag -i "<path to audio file>" -f "music_tagger" -t 0.5

-i, --input         File path to the audio file.
-m, --model         File name of the trained model. Defaults to 'music_tagger.pt'.
-t, --threshold     Probability threshold for tag prediction. Defaults to 0.5.

License

The code in this project is licensed under the MIT License.

Acknowledgements

The project is based on the course Intelligent Audio and Music Analysis held at the TU Wien.