Ultimate Vocal Remover, online

The models the desktop app runs, in a browser tab — no install, no Python, no GPU of your own.

Drop your song here

or click to choose · mp3, wav, flac, m4a · up to 50 MB

What do you want?

VocalsTake the singing out, or keep only the singing.You get 2 files:VocalsSong without vocalsDrumsGet the drums alone, or play along without them.You get 2 files:DrumsSong without drumsBassGet the bassline alone, or play along without it.You get 2 files:BassSong without bassEverythingEvery part on its own, and the song minus each part.You get 8 files:VocalsDrumsBassOther (guitars, keys, synths)Song without vocalsSong without drumsSong without bassSong without the rest

Ultimate Vocal Remover is a free, open-source desktop application, and it is genuinely excellent. It is also a desktop application: you download it, you install it, you fetch model files, and separation runs on whatever hardware you happen to own. On a laptop without a discrete GPU that means minutes per song.

This site runs the same class of models as a hosted service. If you already use UVR and like it, nothing here replaces it — this is for the times when installing it is not practical, or when the machine in front of you cannot do the work in reasonable time.

What is actually the same, and what is not

The models are the same family. The two-way voice-and-backing split uses an MDX-Net model from the same lineage UVR ships, and the four-way split uses Meta's Demucs — the same Demucs the desktop app offers in its model list. On a given song, output from either should sound very close.

What is different is choice and control. UVR exposes dozens of models and a wall of parameters: segment sizes, overlap, aggression, ensemble modes, denoising. That flexibility is the reason serious users prefer it, and it is deliberately not reproduced here — this offers four options and sensible settings.

The other difference is where the compute happens. UVR uses your machine; this uses a rented GPU that spins up for the seconds your song takes and then shuts down.

When to use which

Use the desktop app if you want to compare models, run ensembles, tune parameters, or process a large batch of files locally.
Use this if you are on a machine you cannot install software on, if your laptop has no usable GPU, or if you need one or two songs done now.
Use the desktop app if the audio is confidential and you would rather it never left your computer — that is a completely reasonable requirement and no hosted service can match it.
Use this if you want the newer separation quality without maintaining a Python environment and a folder of model checkpoints.

Questions

›Is this made by the UVR project?

No, and it is not affiliated with it. UVR is an independent open-source project; this is a separate service that runs models from the same open ecosystem. Credit for the models and for UVR itself belongs to their authors.

›Which models exactly?

An MDX-Net instrumental model for the two-file split, and htdemucs — Meta's Demucs v4 — for the four-part split. Both are open and widely used.

›Can I choose a different model?

Not currently. The trade for not installing anything is that the model choice is made for you. If model selection matters to your work, the desktop app is the better tool and that is not a close call.

›Does it upload my audio somewhere?

Yes — separation runs on a server, so the file leaves your machine. It is deleted as soon as processing finishes, and the separated files go when your download links expire. If that is unacceptable for your material, run UVR locally instead.

Where it struggles

Anyone who wants fine control will find this restrictive by design, and should use the desktop application. There are no ensembles, no model comparison, no parameter tuning, and no batch processing — four presets and a daily limit. It is a convenience layer, not a replacement for a real audio workstation tool.