Remove the vocals.
Nothing leaves your browser.
Drop any song and get a clean vocal track and a instrumental, separated locally by a neural network. No upload. No account. Free.
Drop an audio file here
or click to browse · MP3, WAV, FLAC, M4A · up to ~10 min
first run downloads the model once (~67 MB), then it's cached
Step 1
Decode
Your file is decoded to raw audio by the browser itself. It never touches a server.
Step 2
Separate
A vocal-specialist Demucs network runs on your GPU via WebGPU (or CPU as fallback) and lifts the voice out of the mix.
Step 3
Export
Preview both stems and download studio-quality WAV files, straight from memory.
Preparing…
0%
Your stems.
That didn't work
About removing vocals from a song
A vocal remover separates the singing from everything else in a finished recording. Older tools did this with phase cancellation, which only worked when the vocal sat dead centre in the stereo image and usually took the bass and snare with it. This one uses a trained separation model instead, so it can tell a voice from a guitar even when they share the same frequencies.
The result is two files: the isolated vocal, and the backing track with the voice taken out. How clean the split sounds depends on the song. Sparse arrangements separate almost perfectly. Dense mixes with heavy reverb, layered harmonies or loud backing vocals are harder, and you may hear traces of the original vocal in the instrumental.
Everything runs on your own machine. The model is downloaded once and cached, after which the page works offline. Your audio is never sent anywhere.
Also on this site
Privacy promise
Everything happens on your device. Your audio is never uploaded, stored, or listened to by anyone. The site works offline once the model is cached. You can verify this in your browser's network tab: after the model download, no data leaves this page.