Ultimate Vocal Remover
A desktop app that uses neural-network models to pull vocals, instrumentals, and other stems out of a finished mix, on your own machine.
- Type Stems
- Difficulty Easy
- Hardware 64-bit Windows 10 or later, or macOS Big Sur or later, with official bundles that already include Python and PyTorch. Linux instructions are provided for Debian- and Arch-based systems (Medium if you install that way). GPU conversion: an Nvidia GTX 1060 6 GB is the stated minimum; 8 GB of VRAM or more is recommended. AMD support is limited; the project points to a separate AMD branch. Intel Pentium and Celeron systems are not guaranteed. The Windows installer should be placed on the main C: drive.
- License MIT (GUI code)
What it does
Ultimate Vocal Remover GUI (UVR) is an open-source desktop interface for source-separation models. Core developers are Anjok07 and aufr33. The project says they trained the models shipped with the app, except for the Demucs v3 and v4 four-stem models.
You load an audio file and choose a model and output type — vocals, instrumental, or other stems depending on the model. Results vary by mix and by model. This is a tool for practice, remixes, and isolating a vocal or beat you already have the rights to use, not a clear-any-song button.
Why it matters for hip-hop
Stem work is everyday work: chopping an acapella, flipping an instrumental, or checking how a mix is balanced. UVR runs locally, so the file stays on your computer. Used next to a DAW, it is one of the faster ways to get a usable vocal or inst from a stereo bounce when you do not have the original session.
If you are sampling a record you do not own, separating the stems does not give you a license. It only gives you files.
Getting it running
On Windows and Mac, the official path is the installer or disk image from the GitHub releases page. The README says those bundles include what you need. There are separate Mac builds for Apple Silicon and Intel. A DirectML Windows build is offered for some AMD Radeon and Intel Arc cards.
Linux takes more work (Python and dependencies), which is why Linux is Medium. Conversion time depends on your CPU or GPU. Memory errors can often be eased by lowering segment or window size, per the project's troubleshooting notes.
License notes
The UVR GUI code is MIT-licensed. The project asks third-party developers who use the models to give credit. MIT is a permissive software license; it is not a license to the music you run through the app. The bundled Demucs models come from a different project, and Demucs itself warns that the license on some of its weights is unclear.