Vocal Remover takes a song and gives you back the music without the singer — a backing track for karaoke, singing practice or a cover. It uses Demucs, an open-source source-separation model from Meta's AI research, which tells the voice apart from the instruments much more cleanly than the old trick of cancelling the centre of the stereo image.
This runs in the cloud, not in your browser: the audio is sent to a GPU service that runs the model. It needs an account and is a Plus tool — included in Plus and Pro, and free accounts can try it once a day; each run uses 4 credits. Files larger than 4 MB are first re-encoded in your browser to a 96 kbps MP3 so the upload stays small, which costs a little fidelity.
How to use
Sign in and choose a song file — MP3, M4A, WAV or another common audio format.
Start the run and keep the tab open; separating a typical song takes a minute or two.
Listen to the instrumental in the page, then download it.
FAQ
Do I also get the vocals on their own?
No — this tool returns only the instrumental. If you want the isolated vocal track as well, use Stem Separation, which splits the song into vocals, drums, bass and everything else.
Why can I still hear traces of the voice?
Reverb, echo, backing vocals and harmonies are blended into the music in the mix, and no separation model removes them completely. Clean studio recordings separate best; live recordings and heavily compressed files leave more residue.
What does it cost?
It is a Plus tool: included in Plus and Pro, and free accounts get one try a day. Each run also uses 4 AI credits. If the job fails, the credits are refunded automatically.
Where is my song sent?
To the cloud provider that runs the separation model, over an encrypted connection — the separation needs a GPU, so it cannot run in your browser. Only use it for audio you are comfortable processing in the cloud.