Split a song into vocals, drums, bass and the remaining instruments
With adefossez/demucs, MIT (code); the upstream maintainer does not specify clear redistribution rights for HTDemucs.. It runs once on your files and hands you the result.
Open in UGTUGT opens on your computer and shows what it will do. Nothing runs until you press Run.
- It takes
- an audio file with music or singing
inputs/song
- an audio file with music or singing
- It gives you
- vocals
outputs/vocals.wav - drums
outputs/drums.wav - bass
outputs/bass.wav - the remaining instruments
outputs/other.wav
- vocals
- It can reach
- The internet. Downloads the 84 MB pretrained model from the author on first use. The container that reads the song has no network.
- Tested
- On clean amd64 and arm64 machines, 2026-11-04, its check passing on its sample and failing on broken results.
- Its files
- 7 files in UGT's catalogue ↗, revision
47e08e0f6ee8, from829d21890023
No UGT yet?
Set it up once. On a Mac or Linux, in Terminal:
curl -fsSL https://usegithubthing.com/install | shOn Windows, in PowerShell:
irm https://usegithubthing.com/install.ps1 | iex