UVR5 GUI (MDX, VR, Demucs 2-4 and UVR team models)
GUI provides more functionalities and models/AIs compared to Colabs, incl. custom model import:
https://github.com/Anjok07/ultimatevocalremovergui/releases
Official app Win 11 installation tutorial:
MacOS build:
https://github.com/Anjok07/ultimatevocalremovergui/releases/tag/v5.6 (or beta)
MacOS Catalina tutorial (outdated at this point):
https://www.youtube.com/watch?v=u8faZW7mzYs
(you better don’t run Windows build in W10 VM or you will get like 3 hours processing time)
Windows 7 users:
"To use the newest python 3.8+ with Windows 7 install VxKex API extensions and in case of problems select Windows 10 compatibility in EXE installer properties."
Here you can find a searchable PDF guide by the devs for UVR5 GUI describing functions and parameters (can be outdated):
https://drive.google.com/file/d/1RtMRj8FpSpMHlK1XxaBrKoWQlmMfCSmy/view?usp=drive_link
Video guide:
Online guide:
https://multimedia.easeus.com/ai-article/how-to-use-ultimate-vocal-remover.html
(their instructions for installing and using stable version seems to be fine, despite the fact they recommend to clone repo for Macs. It's already available as appropriate binaries for M1 and Intel CPUs (/wo Roformer beta patch at the moment)
Some of UVR5 GUI models described in this guide can be downloaded via the expansion pack:
VIP models
https://www.buymeacoffee.com/uvr5/vip-model-download-instructions
(some older) settings for the GUI:
https://photos.app.goo.gl/EUNMxm1XwnjMHKmW6
(though it's mostly outdated).
(no longer necessary as UVR now has separate DirectML branch and executable:)
Optional fork of UVR GUI for AMD and Intel cards, currently supporting only VR Architecture and MDX using DirectML (Demucs currently not supported). If you have Nvidia card, then use official app above since CUDA is supposed to be faster.
“A four minute and 20 second audio takes about 30 seconds (including saving) using 1_HP-UVR on an Intel Arc A770 16GB. It takes up approximately 6GB of VRAM.”
If you only use MDX models, in most cases it won't be faster than processing with CPU - i5 4460 has similar performance to RX 6700 XT here, so better stick to official app.
Compared to Roformer beta #8, it’s still much faster at least for VR models, but you might get some issues with MDX-Net models, though.
https://github.com/Aloereed/ultimatevocalremovergui-directml
Python command line fork of UVR 5 with current models support:
https://github.com/nomadkaraoke/python-audio-separator
(moves from: https://github.com/karaokenerds/python-audio-separator)
It was based on some outdated UVR code, but probably got updated since then to support more models.