Ultimate Vocal Remover, usually called UVR, is the best free stem splitter if you have a computer. It is open source, it runs the same families of separation models that many paid services use, and because everything happens on your machine there are no accounts, uploads or daily limits. This guide takes you from download to finished stems, and explains which model to use for which job.
We make Autunes, which has a paid six-stem splitter of its own. If you only split songs now and then and have a laptop, UVR is the better deal, and we would rather tell you that than not.
What you need
- A Windows, Mac or Linux computer. A graphics card speeds things up a lot but is not required.
- A few gigabytes of free disk space for the app and its models.
- The best-quality copy of the song you have: WAV or a high-bitrate MP3 separates far more cleanly than audio ripped from a video.
Step 1: download and install
- Go to the official Ultimate Vocal Remover GitHub page and open Releases. Avoid copies on other download sites.
- Windows: download the installer and run it.
- Mac: download the build for your chip (Apple Silicon or Intel). The first time you open it, macOS may block it; allow it under System Settings → Privacy & Security.
- Open UVR. The first launch can take a minute.
Step 2: choose the input and output
Click Select Input and choose your song; you can pick several to process them in a batch. Click Select Output and choose a folder for the stems. Leave the output format as WAV for the best quality; switch to MP3 only if disk space is tight.
Step 3: pick the right model for the job
UVR offers several process methods, each with its own models. The names change between versions, but the families stay the same:
- Vocals and instrumental (karaoke, acapella): choose MDX-Net or a RoFormer vocal model. These give the cleanest two-way split.
- Drums, bass, other and vocals (four stems): choose Demucs and a recent htdemucs model.
- Six stems including guitar and piano: some Demucs models add guitar and piano; quality on those two varies by song.
- Removing echo or reverb from a vocal: run a de-reverb model on the vocal stem as a second pass.
Step 4: process
Tick GPU Conversion if you have a supported graphics card. Press Start Processing. UVR downloads the model the first time you use it, then separates the song. On a laptop without a graphics card a four-minute song can take several minutes; with a decent card it takes well under a minute. The stems appear in your output folder named after the song and the stem.
Fixing common problems
- Music left in the vocal: try a different vocal model, or run the vocal stem through a second model.
- Vocal left in the instrumental: use an instrumental-focused MDX-Net model, or ensemble two models.
- Very slow: turn on GPU Conversion if you can, close other apps, or process shorter files.
- Out-of-memory errors: lower the segment size in the model settings.
- Thin or watery sound: that is bleed removal working too hard; try a model with a lower aggression setting.
When UVR is not the right tool
UVR is ideal for the occasional split on a computer. It is less convenient if you only have a phone, if you split many songs every week and do not want your laptop busy, or if the split is just the first step and you want to keep working on the song: add an instrument, remix it into another style or re-sing a line.
For phones, BandLab Splitter and the Moises free plan are the easiest free options, with limits on songs or stems. For the keep-working case, Autunes Stems splits a song into six parts (vocals, drums, bass, guitar, piano, other) in about a minute, inside the same Studio where you can layer, remix and edit. It costs 5 credits a song on Pro or Creator and is not on the free plan. Our free stem splitter page compares all of these side by side.
A word on rights
Splitting a song for practice at home is fine; publishing its acapella or karaoke track needs permission from whoever owns the recording.