Stemgen: Stem file generator

Hello,

I just released a script that convert any track into a Stem file: https://github.com/axeldelafosse/stemgen

Create your own stems with only one command:

$ ./stemgen -i track.wav

Have fun!

 

I downloaded the archive, could you explain how to use it?

 

Unfortunately this script is not easy to use. Sorry!

We are working to make Spleeter output a Stem file compatible with Traktor directly so you won't need this script. Spleeter is designed to be easy to use. I'll let you know when it's ready.

 

Awesome work, thank you so much. Works just fine, all the tags seems to be there except maybe the Comment one? No idea why

I made a dirty edit to use demucs instead, and also just added the ouput folder to my music folders imported at startup, so it analyse straight away, I can churn lossless stem as fast as my cpu can handle encoding them.

Of course it would be nice as a drag and drop batch app. But thanks again!

 

Hey Nico, happy to hear that!

Yeah we don't copy all the tags. I can add `comment` if you want.

Feel free to submit a pull request to add some options (using spleeter or demucs, specifying the output folder, batching, etc...), I'll take the time to review and merge.

I'm curious, why demucs? Is it better?

I'd love a GUI too, I will see what I can do.

 

Don't even know what a pull request is, I'm figuring it all as I go But I'll look. Personally I use comment and comment2 a lot, batching would be nice, like put all the tracks you want done in a folder would be convenient, it would be able to run overnight. Output folder was fine for me once I've seen where it was. Switching from spleeter to demucs doesn't look simple where I stand, because I have to launch the proper anaconda environment first in a terminal window first and then I can run either your app or the slightly edited version I did that recover the splited wavs from demucs output and put them where the rest of the script expect them.

As to why, I don't like the phasey sound spleeter has, demucs has other artifacts but I find them less annoying and I do like the end result better than spleeter yes.

Not necessarily a GUI, because there isn't much to configure after all, but a droplet would be nice, like you just need to pull in the Finder the files you want on it and it starts the conversion. A self contained app that doesn't need to install the whole python environment and everything would be perfect for the average end user I guess, but I don't know how complicated that would be to make.

 

Haha no worries. I added comment and comment2 here.

Mmh, interesting... You're right, a droplet would be perfect!

 

Oh that's sweet, thanks!

For design ideas, you can maybe take a look at XLD, it's an audio converter on the droplet principle, you configure the type of conversion you want in the prefs, it does batching with progress bars, etc, it's quite nifty. Also it's open source so I guess you may be able to reuse parts of the code, idk.
https://tmkk.undo.jp/xld/index_e.html

As far as configuration, one thing I think people may want is lossy/lossless stems as the lossless version is quite big.

 

does this work on Windows or on Mac only