Vocal remover
Upload a song and get two tracks back: the vocals on their own and the music without them. An AI model trained for music separation does the work on our servers, as MP3 at 320 kbps or lossless WAV. Free, no account, files up to 10 minutes.
How it works
- Choose a song (MP3, WAV, FLAC, M4A, OGG, or a video; up to 500 MB and 10 minutes).
- Pick what you want back: vocals and instrumental, four stems, or just one of them, and MP3 or WAV.
- Listen in the page and download. Everything is deleted 60 minutes after the job started.
Questions
How clean is the separation?
Good on most studio recordings, but not perfect: reverb tails, backing vocals and instruments that sit in the vocal range (a lead guitar, a sax) can leave traces in the wrong track, and heavily compressed or low-bitrate files separate less cleanly. Upload the best-quality copy you have.
How long does it take?
A few minutes for a typical song: the model runs on CPU servers and the progress bar shows how far along it is. Longer files take proportionally longer.
What happens to my file?
It is uploaded to our server, separated there, and the upload and every result are deleted automatically 60 minutes after the job started. Nothing is kept, nothing is used to train anything, and nobody listens to it.
What are the limits?
Files up to 500 MB and 10 minutes long. One song is separated at a time, so at busy moments yours waits in a short queue; the page shows how many are ahead.
Which model does the separating?
Hybrid Transformer Demucs (htdemucs), the open-source music source separation model published by Meta AI researchers under the MIT licence. It runs on our own servers; your audio is not sent to any other service.
May I use the result?
The separated tracks are still the song. Use them the way you are allowed to use the original: practice, study, remixes for yourself, or material you own or have licensed.
Do I need an account?
No. There is nothing to sign up for and nothing to install.