xavriley/demucs_v3_saxophone_separation

Saxophone Source Separation model

• paper: https://arxiv.org/abs/2405.16687

Source separation model trained for the paper:

Reconstructing the Charlie Parker Omnibook using an audio-to-score automatic transcription pipeline Xavier Riley and Simon Dixon SMC 2024, Porto, Portugal

Separates tenor sax cleanly but struggles a little with high register playing.

Dependencies

pip install demucs==3.0.6

Usage

python -mdemucs --repo . -n filosax_demucs_v3_14.22_SDR.th -o . input.wav

Alternative

For an alternative you could also check out the UVR Wind model

pip install audio-separator[gpu] # or leave off [gpu] if you only have CPU
audio-separator -m 17_HP-Wind_Inst-UVR.pth input.wav