Approximate separation with a local model
Choose an audio file and a clip up to 60 seconds. A self-hosted UVR model produces vocal and accompaniment WAV tracks.
Example and limitations
Results depend on the recording and device. Vocal/music leakage can remain. This is not forensic evidence or guaranteed studio quality. Cancel if performance is unsuitable.
Does it work on every phone?
It depends on memory and WebAssembly support. Low-memory devices may need a shorter clip or a computer.