Demucs

Package

Open-source model that splits songs into vocals, drums, bass and other stems

Price: Free, open source
Access: None, runs locally

About

Hybrid Transformer source-separation model and Python package, MIT-licensed. Splits a track into vocals, drums, bass and other, or two stems for karaoke; an experimental 6-stem model adds guitar and piano (piano bleeds). Runs on CPU or GPU. Maintained, but its author adds no new features.

What you can do with it

Split a song into vocals, drums, bass and other stems on your own machine
Remove the vocals to make a karaoke or instrumental track with --two-stems
Batch-separate audio files from Python or the command line, on CPU or GPU

Get started

Install with pip install -U demucs, or run it with uvx demucs
Install ffmpeg for FLAC output and formats the decoder can't read
Add -d cpu, or a smaller --segment, if GPU memory runs out

Example

python3 -m pip install -U demucs

# Four stems: vocals, drums, bass, other
demucs myfile.mp3
# Vocals plus accompaniment, for karaoke
demucs --two-stems=vocals myfile.mp3
# Six stems, adding guitar and piano
demucs -n htdemucs_6s myfile.mp3

Details

Hosting: Runs locally
Available in: Worldwide
Official SDKs: Python
MCP server: None