Demucs

Demucs separates mixed music recordings into parts for musicians who want to work with individual instruments or vocals. The author maintains a separate repository, with no active feature development and slow responses expected.

• Separate stems: Extract vocals, drums and bass alongside the remaining accompaniment from a song.
• Separation models: Use Hybrid Transformer Demucs or classic Hybrid Demucs to separate tracks with different model architectures.
• Desktop interfaces: Demucs-Gui and Ultimate Vocal Remover support Demucs for people who prefer a graphical interface.

The Python software is open source under the MIT license. It runs locally on Windows, macOS and Linux, and a community Docker image definition is available. Processing can use a CPU or CUDA GPU. GPU processing with the default settings needs about 7 GB of GPU memory; smaller audio segments reduce memory use but can affect separation quality.

Local separation processes recordings on your computer. Google Colab and a Hugging Face Spaces demo offer remote alternatives, where audio goes to the hosted service for processing.

This listing links to Alexandre Défossez’s maintained repository, which distributes Demucs v4 and Hybrid Transformer checkpoints under MIT. The author is no longer actively developing new features and warns that responses may be slow. The archived Meta repository is the earlier source, not the current maintenance location.