STEM / AVAILABLE NOW

Vocal Remover

Separate voice and music. Create instrumental and vocal stems using an on-device separation model.

Local AI modelOne-time 172 MB model downloadNo signup
STEM / dedicated toolOne-time 172 MB model download
VOCALSAwaiting separation
INSTRUMENTALAwaiting separation
GUIDE / vocal remover online

How does a vocal remover online separate a finished mix?

A vocal remover online estimates the vocal source and remaining accompaniment in a finished mix. SoundTools runs a four-source Demucs model through local WebGPU, exposes its vocals prediction, combines drums, bass, and other sources into an instrumental WAV, and provides separate previews without uploading the selected song.

RELATED USES
  • voice remover
  • remove instruments
  • isolate vocals
  • vocals only
FIELD NOTES / 01

Expect two estimates, not the original studio stems

The first separation downloads and caches an approximately 172 MB Demucs model. A supported desktop Chrome or Edge browser then resamples the decoded mix to 44.1 kHz stereo and processes overlapping chunks through WebGPU. Keep the tab open until both the Vocals and Instrumental players appear; model download and GPU work can take longer than a simple audio effect.

Demucs predicts vocals, drums, bass, and other sources. This page exports the vocal prediction directly and combines drums, bass, and other into one instrumental WAV. A centered, dry lead vocal usually separates more cleanly than stacked harmonies, loud reverb, distorted guitars, or instruments that occupy similar frequencies.

REFERENCE / MODE CHOICE

Vocal remover vs four-stem splitter vs center cancellation

The output count and separation method determine what you can do after processing.

MethodBest forWhat it createsImportant limit
Two-output vocal removerVocals-only and karaoke-style instrumental resultsOne vocal prediction plus drums, bass, and other combined as instrumentalYou cannot adjust drums, bass, and other separately
Four-stem splitterArrangement study and more flexible authorized remixingSeparate vocals, drums, bass, and other WAV predictionsThe same model bleed remains, with more files to review
Center cancellationSimple older stereo mixes with centered vocalsSubtracts shared left/right content rather than using a source modelCan remove centered drums and bass while leaving stereo vocals or reverb
PROCESS / THREE STEPS

How to use the vocal remover online

  1. 01

    Open a full music mix

    Choose a song you own or may process. The browser decodes it locally and shows the original player before any model download begins.

  2. 02

    Start WebGPU separation

    Use a current supported desktop browser, allow the approximately 172 MB model to download on the first run, and keep the tab open while overlapping chunks are processed.

  3. 03

    Preview both estimated stems

    Listen to Vocals and Instrumental separately for bleed or missing detail, then download either or both 44.1 kHz stereo WAV files.

SIGNAL NOTES / VERIFIED BEHAVIOR

What this tool actually does

Clear limits are part of a useful tool. These values describe the processor currently running in this page.

01 / Model Demucs · about 172 MB

The model downloads only when separation starts and can be reused from browser cache on later runs.

02 / Processing Local WebGPU · 44.1 kHz stereo

The mix is resampled and processed in overlapping local chunks on a supported GPU.

03 / Outputs Vocals + instrumental WAV

The instrumental result combines the predicted drums, bass, and other sources while omitting the predicted vocal stem.

USE CASES / 02

Useful ways to use estimated vocal and instrumental stems

  • Karaoke and rehearsal reference

    Use the instrumental estimate to practice a part while checking for remaining vocal bleed before a performance or private session.

  • Vocal listening and transcription

    Use the vocal estimate to hear lyrics, phrasing, breaths, or harmonies more clearly in a mix you may analyze.

  • Authorized remix preparation

    Use both outputs as editable starting material, understanding that they are estimates rather than the original multitrack recording.

QUESTIONS / PRACTICAL ANSWERS

Questions about this tool

Answers based on the current browser processor—not promises about a future version.

01Can I remove vocals and keep the instrumental?

Yes. Download the Instrumental WAV, which combines the predicted drums, bass, and other sources while omitting the predicted vocal stem.

02Can I remove instruments and keep only vocals?

Yes. Download the Vocals WAV to keep the model’s vocal estimate. Instruments that overlap the voice may still bleed into that result.

03Why do I still hear vocals in the instrumental?

Reverb, backing vocals, distortion, and overlapping frequencies can be assigned partly to other sources. Separation is a model estimate rather than access to the original studio tracks.

04Which browser and hardware do I need?

Use a current Chrome or Edge browser on a desktop device with supported WebGPU. The workbench reports when a compatible high-performance adapter is unavailable.

05How large is the vocal-remover model?

The Demucs model is approximately 172 MB. It downloads only after separation starts and may be reused from the browser cache on later visits.

06What format are the separated stems?

The tool creates 44.1 kHz stereo WAV files for Vocals and Instrumental so each result can be previewed and downloaded independently.

07Is the song uploaded during separation?

No. The model file is downloaded to the browser, but the selected mix is decoded, resampled, processed, previewed, and exported on the current device.

08May I publish or remix the separated audio?

Only when you own the source or have permission from the relevant rights holders. Source separation changes the audio but not its copyright or license.