Audio separator (UVR / MDX / Demucs)
AudioOfflineGPU optionalSome models non-commercial
High-quality vocal/music separation using UVR, MDX-Net, BS-Roformer, and Demucs models. Quality > the built-in ffmpeg mid/side trick at the cost of a one-shot model download (~50-500 MB) and a few seconds of inference per minute.
Install it from Subtitld
Settings Add-ons audio-separator (UVR / MDX / Demucs) Install
Subtitld downloads it, checks its SHA-256, and keeps it updated.
- Models
- 5
- Add-on download
- 295–432 MB
- RAM, at least
- 4 GB
Models
| Model | Download |
|---|---|
| Kim_Inst.onnx Kim_Inst.onnx | |
| UVR-MDX-NET-Inst_HQ_3.onnx UVR-MDX-NET-Inst_HQ_3.onnx | |
| UVR_MDXNET_KARA_2.onnx UVR_MDXNET_KARA_2.onnx | |
| model_bs_roformer_ep_317_sdr_12.9755.ckpt | |
| htdemucs_ft.yaml |
Options
In Settings › Add-ons › audio-separator (UVR / MDX / Demucs) › Configure.
Separation model
Default:
Kim_Inst.onnxInference device
Default:
cpuOutput format
Default:
FLACPlatforms
| System | Download |
|---|---|
| Linux x86-64 | 432.4 MB |
| macOS (universal) | 312.2 MB |
| Windows x86-64 | 295.5 MB |
License
Add-on
MIT (wrapper) + MIT (python-audio-separator) + per-model (see model card)
MIT (wrapper) + MIT (python-audio-separator) + per-model licenses (some UVR/MDX checkpoints are CC-BY-NC for non-commercial use)
Versions
- 0.0.5
Automated release for audio-separator v0.0.5.