Demucs
State-of-the-art deep-learning music source separation (vocals, drums, bass, other) by Meta.
| Language | Python |
|---|---|
| Category | ML & generative |
| License | MIT |
| Platforms | macOS Windows Linux |
| Install | pip install demucs |
| First released | 2019 |
| Maintained | Not actively maintained (last activity 2023) |
| Links | Source |
Strengths
- Top-tier separation quality via Hybrid Transformer models
- Multiple pretrained models with a simple CLI
- Supports 4- and 6-stem separation
Limitations
- Repo archived (2024); no active development
- GPU strongly recommended, slow on CPU
Best for
High-quality offline stem separation for remixing, karaoke, or sampling.
Minimal example
import demucs.separate
demucs.separate.main(["mytrack.mp3"])