Dzzz

Vocal Remover

On-device

Separation runs on your device. The one-time model download comes from huggingface.co; nothing else leaves your device.

Split a song into an instrumental track and an isolated vocal track for free — the Spleeter separation model runs entirely on your device after a one-time download from huggingface.co, so the audio itself never leaves it. Both results lose sound above about 11 kHz, so cymbals and "air" sound dull, and this is a good separation with audible artefacts rather than a studio-quality stem. No uploads, no account.

About this tool

What you provide

  • A song or audio clip in any format your device can decode.

What you get

  • An instrumental WAV track with the vocals removed.
  • An isolated vocal WAV track, plus an audio player to preview both before you download them.

Limits

  • Both results lose sound above about 11 kHz, so cymbals and "air" come out dull — a limit of the separation model itself.
  • A good separation with audible artefacts, not a studio-quality stem — vocals are not removed perfectly.
  • Needs a one-time ~38 MB model download from huggingface.co before the first run; the separation itself runs on your device on your device.
  • Processing takes roughly as long as the track itself, so a long song can take several minutes and depends on your device's memory.

How to use it

  1. Add a song

    Drop an audio file on the dropzone, or tap it to choose one from your device.

  2. Allow the one-time model download

    The first run needs a ~38 MB Spleeter model from huggingface.co; tap to download it once, and it stays cached for later visits.

  3. Separate the track

    Tap Separate vocals; the split is processed on your device, with progress shown as parts complete and a Cancel button if you change your mind.

  4. Preview and download both stems

    Play the instrumental and vocal tracks in the built-in players, then download whichever one you need as a WAV file.

Frequently asked questions

Is Vocal Remover free to use?

Yes, Vocal Remover is free and splits a song into an instrumental track and a vocal track right on your device.

Does it upload my song to a server?

No — separation runs on your device. The only network request is a one-time model download from huggingface.co; the audio itself is never sent anywhere.

Will the vocals be removed perfectly?

No. It's a good separation with audible artefacts, not a studio-quality stem, and both results lose sound above about 11 kHz, so cymbals and "air" come out dull.

How long does it take?

The tool measures it rather than guessing. It times a short benchmark on your device the first time, then uses your own past runs, and shows a range that narrows as the separation progresses. Speed depends heavily on the device, so a figure quoted here would be wrong for most people.