Open-Unmix
Open-source deep learning reference for music source separation.
About
Open-Unmix is an open-source reference implementation for music source separation, developed by Fabian-Robert Stoter and Antoine Liutkus at Inria in collaboration with Sony. It uses bidirectional LSTM networks to split a mixed track into vocals, drums, bass, and a residual other stem.
It was built to give researchers a reproducible baseline and to give artists a free tool with pretrained models. Implementations exist for PyTorch and other frameworks, and the models can be loaded in one line through torch.hub.
What it can do
- Four-stem separation into vocals, drums, bass, and other
- Pretrained UMX, UMXHQ, and UMXL models
- PyTorch reference implementation
- One-line loading through torch.hub
Setup
Install with pip install openunmix, fetch the pretrained weights, then run the umx command on a WAV file or call separate() from Python.
Prerequisites
Python 3 and PyTorch.
Use cases
Works with
Trust score — 7/10
Scored 7/10 (metrics tier, metrics-v1). Evidence: Verified via vendor site/docs/GitHub and reputable coverage, wave 131; Trending score 70/100 from researcher assessment.
Sources
Every fact above was checked against these sources. Where a claim comes from the provider and couldn't be independently verified, the text says so.