A desktop app that un-sings your favorite songs

What it does

Ultimate Vocal Remover GUI is a cross-platform desktop application that strips vocals from audio files using source-separation neural networks. The core developers trained most of the bundled models themselves; only the Demucs v3 and v4 4-stem models come from elsewhere. It also packs in time-stretching and pitch-shifting via Rubber Band, though you’ll need to bring your own FFmpeg for non-WAV files.

The interesting bit

This isn’t a thin wrapper around someone else’s API. The team trained custom models, ships prebuilt installers bundling Python and PyTorch, and maintains a separate branch for AMD GPU support via DirectML. That’s an unusual amount of infrastructure for a music utility.

Key highlights

• Runs on Windows 10+, macOS Big Sur+, and Debian/Arch-based Linux (64-bit only)
• GPU acceleration via CUDA on Nvidia cards (GTX 1060 6GB minimum, 8GB VRAM recommended); MPS support for Apple Silicon
• AMD Radeon and Intel Arc support exists but is explicitly labeled limited
• Remembers settings between sessions; model load and I/O performance were improved in recent releases
• First launch on Mac can take 5–10 minutes while it sorts itself out

Caveats

• Must install to the C:\ drive on Windows; secondary drives cause instability
• Intel Pentium, Celeron, older Macs, and macOS Catalina or below are not guaranteed to work
• Memory allocation errors may require manually lowering segment or window sizes

Verdict

Worth a look if you produce karaoke tracks, remixes, or need clean instrumental stems and don’t want to pay for a cloud service. Skip it if you’re on budget hardware or expect a polished, bug-free experience out of the box.