This project is Python Based, and has an alarming amount of signs pointing to it being vibe-coded. It mentions multiple appimages yet only 1 exists, only has one commit for all of the code, the code of conduct being, word for word “This project adheres to a code of conduct. By participating, you are expected to uphold this code. Please report unacceptable behavior to the project maintainer,” the readme doesn’t make any sense in some areas, and not only will the appimage not run on my machine, the build instructions and build scripts don’t work and sometimes point to files that don’t exist.
A.I. Stem seperation is a good use of the tech, but this just looks like vibe-coded nonsense.
If the project wasn’t “vibe coded”, would that make it not working for you feel any better? You haven’t reported any bugs on the github. It works fine for me, so this is an issue for you rather than the program per se.
AFAIK this is the only open source GUI on Linux available for using Stem Separation technology so far. If there is a problem other than it not personally working for you, do say but just deriding something as “vibe coded nonsense” is not helpful to anyone.
If it wasn’t vibe coded, I would’ve in fact reported the issue’s I’ve seen to the issue tracker, like referencing links and files that don’t exist. But I suppose I’ll take the time to make the issues on the project.
Aside from the fact that the “Originality Statement” link is 404, the fact that it’s so adamant about not being derived from UVR (aka ultimatevocalremover) is interesting when you look at some of the change history like this:
I think most of us around here appreciate open source. Personally, I’ve found 3D prints of open designs I’ve published on Thingiverse for sale on e-bay. I find that obnoxious. Don’t think I’ll be buying the StemWeaver “developer” a coffee any time soon.
What underlying features and problem mitigations does one seek when using a stem separation tool? The ones I recall from my own usage is the quality, speed of getting the stems, number of stems, and regarding problem mitigations: looking to avoid the drums bleeding into the wrong stem output. (I hope that’s the right usage of the word bleeding).
Also slightly off topic, a part of me wonders what quality of stem separation can be achieved without any AI/training of models, and whether there’s any “real-time” need/use for stem separation. (Calling it off-topic because I assume it likely doesn’t exist, or is too rare/tedious to make.)
Meanwhile, stem separation has become instrumental in helping people learn to play music, make covers, etc. Anyone who has seen Chad Smith drumming the shit out of a 30 seconds to mars song at Drumeo YT channel can attest to that.
