Open-Unmix
github.com
- Free plan
- Windows
- macOS
- Linux
- Self-hosted
What Open-Unmix is
Open-Unmix is a free PyTorch reference implementation for separating music sources, aimed at researchers, audio engineers and artists. Its ready-to-use models split pop music into vocals, drums, bass and other instruments, while the umxse model handles speech enhancement and was trained on the 28-speaker Voicebank+DEMAND corpus. Pre-trained options include umxl, umxhq, umx and umxse, with spectrogram variants also available. The project is designed to help researchers reproduce results, explore model architectures and add their own data. Each source model uses a three-layer bidirectional LSTM; the project says this design cannot run online or in real time. The Python package installs with pip, and a Docker container is also available. Models can be loaded from Python or torch.hub, and a Google Colab notebook supports online experimentation without local setup. The command line accepts WAV, FLAC and OGG, but not MP3; stempeg can extend input and output options, including STEMS export. The software is MIT-licensed, while umxl weights carry a non-commercial CC BY-NC-SA 4.0 license.
Who it suits
It suits researchers, audio engineers and artists who want to separate musical stems or explore source-separation models. Users looking to work with umxl weights should note their non-commercial license.
Good to know, and what to watch
- Ready models separate vocals, drums, bass and other instruments
- Includes an umxse speech enhancement model
- Python, torch.hub, Docker and Colab options
- Open-source MIT software
- Model design cannot run online or in real time
- Command line does not support MP3
- umxl weights are licensed for non-commercial use
Plans
Every plan and priceOpen-Unmix
Open source
The verdict
Open-Unmix offers pre-trained source-separation models alongside code for research and extension. Its real-time limitation, command-line format support and separate umxl weights license are important to account for.
What the maker publishes
- Audio formats
- The command line supports WAV, FLAC, and OGG files but not MP3; stempeg can expand supported input and output formats, including STEMS export. source
- Installation
- The Python package can be installed with pip, and the project also provides a Docker container. source
- Integrations
- Pre-trained models can be loaded from Python or through torch.hub, and a Google Colab notebook offers online experimentation without local setup. source
- Maker
- The SigSep GitHub organization describes itself as providing open resources for audio source separation. source
- Model architecture
- Each source model uses a three-layer bidirectional LSTM, and the project states that this design cannot run online or in real time. source
- Model options
- Pre-trained options include umxl, umxhq, umx, and umxse, with spectrogram variants also available. source
- Model weights license
- The umxl weights are licensed for non-commercial use under CC BY-NC-SA 4.0. source
- Music stems
- Its ready-to-use models separate pop music into vocals, drums, bass, and other instruments. source
- Purpose
- Open-Unmix is a PyTorch reference implementation for music source separation aimed at researchers, audio engineers, and artists. source
- Security and compliance
- The opened project pages state an MIT license but do not describe security controls or compliance certifications. source
- Software license
- The repository lists the software license as MIT. source
- Speech enhancement
- The project provides an umxse speech enhancement model trained on the 28-speaker Voicebank+DEMAND corpus. source
- Support
- The project directs users to GitHub issues, Gitter chat, and Google Groups for technical help and support. source
- Training and extensibility
- The code is designed to let researchers reproduce results, develop architectures, and add their own data, while model training is documented separately from the package. source
Checked against the maker's pages on 8 Oct 2026.