UNet-pytorch

Pytorch implementation of U-Net for music source separation.

Installing Dependencies

• FFmpeg for audio I/O sudo apt install ffmpeg
• python module requirements

python3 -m venv env
. env/bin/activate
pip install -r requirements.txt

Training

python train.py -v voc_weights.pt -a accompaniment_weights.pt

Testing

python separate_musdb.py -v voc_weights.pt -a accompaniment_weights.pt

Separating songs

python song_separator.py -v voc_weights.pt -a accompaniment_weights.pt -i song.mp3

Use the -h flag for more information on the command line arguments