Skip to content

Stem Splitter

Free Vocal Remover

Take the vocals out of a song for a karaoke or practice track, or keep only the vocals for a remix. Demucs, an open AI model made for separating music, runs on this device's graphics card: your song is never uploaded.

Stem separation

Separate vocals, drums, bass or other instruments with the Demucs model running on this device’s graphics card. Each stem model is about 166 MB and downloads once, the first time you use it; your audio is never uploaded. Separation is thorough rather than instant: expect several minutes per stem on a laptop.

Checking whether this device can run stem separation…

Tracks up to 100 MB and 12 minutes long for this choice.

Cleanup and mix & master

These apply to every stem as it finishes. Change them afterwards and press Apply to rebuild the stems from the separated audio without separating again.

Remove the vocals from a song

  1. Choose your song: an MP3, WAV, FLAC, OGG, M4A or AIFF file, or a video with sound.
  2. Keep Vocals and instrumental selected. It is one model of about 166 MB, downloaded once, and the fastest pair.
  3. Press Separate stems. Each part appears as soon as it is ready, with the time left shown as it works.
  4. Listen to the instrumental and the vocals, then download either one or both, or mix and master a part right here.

Questions

Is the vocal remover free?
Yes. There is no limit on songs, no watermark and no account to make.
Is my song uploaded?
No. The separation runs in your browser on this device, so the music stays with you. The only download is the model, once.
What do I need to run it?
A browser with WebGPU on a device with a graphics card, such as a recent Chrome or Edge on a computer. The page checks this device before you start and tells you if it cannot run the model.
Can it separate the drums and bass too?
Yes. Choose All four stems for vocals, drums, bass and the other instruments: quick with one model, or best with four models of up to about 664 MB in all.
How clean is the result?
Demucs is one of the strongest open models for this, but no separation is perfect: reverb and backing vocals can leave traces in the instrumental. Listen to each part before you use it.

More from Stem Splitter: open the whole workspace.