With CUDA support (Windows & Linux)
pip3 install torch torchvision torchaudio --extra-index-url https://download.pytorch.org/whl/cu116 demucs SoundFile cx-Freeze
CPU only
pip3 install torch torchvision torchaudio demucs SoundFile cx-Freeze
And then:
cxfreeze main.py --target-dir=dist --target-name=demucs-cxfreeze --packages=torch
Copy venv/Lib/site-packages/_soundfile_data
to dist/lib/_soundfile_data
Additional Dependenices (should be shipped with frozen Demucs; not included with this repository's releases)
The directory containing ffmpeg
and ffprobe
binaries should be added to the PATH
environment variable in the environment in which demucs
is run.
https://www.gyan.dev/ffmpeg/builds/ffmpeg-release-essentials.zip
- https://evermeet.cx/ffmpeg/ffmpeg-107374-g046b05082d.zip
- https://evermeet.cx/ffmpeg/ffprobe-107374-g046b05082d.zip
Downloaded models and their corresponding YAML file should be placed in a directory, and that directory should be passed to Demucs via the --repo
command line argument.
- https://dl.fbaipublicfiles.com/demucs/mdx_final/83fc094f-4a16d450.th
- https://dl.fbaipublicfiles.com/demucs/mdx_final/7fd6ef75-a905dd85.th
- https://dl.fbaipublicfiles.com/demucs/mdx_final/14fc6a69-a89dd0ee.th
- https://dl.fbaipublicfiles.com/demucs/mdx_final/464b36d7-e5a9386e.th
- https://raw.githubusercontent.com/facebookresearch/demucs/main/demucs/remote/mdx_extra_q.yaml
https://raw.githubusercontent.com/facebookresearch/demucs/main/demucs/remote/files.txt