Repository navigation
[Bug]: Lead + Backing Vocal Separation still generates a unified stem #832
Description
Before reporting
- I'm on the latest release and the issue still happens.
- I searched existing issues and didn't find a duplicate.
What happened?
I selected Lead + Backing as the Vocal mode for both an ogg and wav file, both resulted in a single vocal track, backgrounds and leads were not separated into different tracks.
Steps to reproduce
- Import song
- Import audio file (ogg or wav)
- Select Lead + Backing under Vocals
- Extract Stems
Operating system
Linux
StemDeck version
v0.20.0
How did you install it?
Linux tar.gz
Logs / screenshots
No response
Activity
MetalMan1245 commentedon Oct 22, 2026
Worth mentioning that I am using the CPU only version, not the Nvidia version, I am not convinced that I have an Nvidia card capable of running the Nvidia version and am in the process of moving so would be unable to try it anyway.
fix(vocal-split): say when Lead + Backing could not be made (#834)
thcp commentedon Oct 22, 2026
The likely cause was the Linux CPU package. It shipped a build of torchvision that the Lead + Backing model cannot load, so the split failed and you got one vocal track. That is fixed in 0.21.0 (see #778), which is a full download.
The app now also tells you when Lead + Backing could not be made, says why, and offers Try again on the same song. It shows the option as unavailable when it cannot run on that install.
If it still happens after you update, look in logs/setup.log for a line starting with WARMUP_FAILED vocal_split, then open a new issue and paste it.
MetalMan1245 commentedon Oct 22, 2026
Yep, still getting the issue after cloning the repo and setting up with the run script. I was unable to find a WARMUP_FAILED line in the only log I have accessible, stemdeck.log. I also have the old log from when I installed v0.20.0, both are attached for clarity.
setup.log
stemdeck.log