This repository is a fork of the official implementation of SCNet: Sparse Compression Network for Music Source Separation.
It moves dependency management to uv and fixes some instability issues during training.
First, you need to install the requirements.
cd SCNet-main
uv syncWe use the accelerate package from Hugging Face for multi-gpu training.
uv run accelerate configYou need to modify the dataset path in the /conf/config.yaml. The dataset folder should contain the train and valid parts.
data:
wav: /path/to/datasetThe training command is as follows. If you do not specify a path, the default path will be used.
uv run accelerate launch -m scnet.train --config_path path/to/config.yaml --save_path path/to/save/checkpoint/