How it works

Your music stays with you

The tools on this site run inside your browser. When you pick a song, it is read and processed by your own phone or computer. It is not uploaded, and there is no account to create.

The site is hosted on Cloudflare, which, like any host, sees ordinary visit data such as the address of the page you opened. Your audio is not part of that.

What the page downloads

To separate the vocals the page needs a model, a file that learned to tell a voice from instruments. It is downloaded the first time and kept in your browser, so the next times start right away. Phones and tablets get a light model (about 40 MB). Computers get a bigger one that separates more cleanly (about 180 MB).

Limits

The result depends on the song. Clean recordings separate better than very busy ones. The vocal track loses the highest treble, and a little of the voice can remain in the instrumental.

Only separate music you have the right to use.

Software this site uses

Piece	What it does	License
ONNX Runtime Web	runs the model in the browser	MIT, Microsoft
Spleeter	the vocal model on phones	Deezer, code under MIT
HTDemucs	the vocal model on computers	Meta, code under MIT
demucs-web	audio math (FFT)	MIT, timcsy
lamejs	saves MP3 files	LGPL-3.0
Bricolage Grotesque	the title font	SIL Open Font License