Alex HallYesterday, I boosted a post with a music stem separator for Windows. I have since tried it, and it's amazing. The interface is simple and basic, and the outputs of my first test were great. It kept a bit of keyboard in the vocals, but that was the only problem, and it has a bunch of models I could try. https://www.dropbox.com/scl/fo/jnicbs8cjc8orukpe8vhv/AA2n8ya1sWTvldMvyslOk_0?rlkey=au1g8jwx30xub8n9b5uks8gv5&e=1&dl=0
More on the stem separator: at least for the default models, it seems to assume that instruments will be present. If your song has guitars, bass, drums, piano, and vocals, it'll do fine. I threw a song at it that had vocals, several horns, drums, and an organ-style keyboard, and it didn't get those at all. I imagine I'd have to choose models meant for such instruments, but there are a LOT of models, and they don't always have helpful names.
@alexhall Hmm what about removing the trumpet and sax from a song like this? I want to play it on my wind controller with just the drums/bass/piano accompaniment.
Eddies theme from who frame roger rabbit
https://www.youtube.com/watch?v=JN_vGOhG-7k
Eddies theme from who frame roger rabbit
https://www.youtube.com/watch?v=JN_vGOhG-7k
Who Framed Roger Rabbit - Eddie's Theme
@jpellis2008 It makes a file for each instrument, plus a copy of the song with no vocals. Currently, it doesn't let you specify one instrument to remove, or multiple instruments to put into one file.
Borris@alexhall @jpellis2008 You would need a specific model trained on those instruments, and enough stems to cover it. AFAIK, none of the general purpose models are that specific.
@BorrisInABox @alexhall That's too bad. Reall y want this one and I can't find a version with just accompaniment, even to purchase.
Andre Louis@Scott
Interesting, maybe we can get @Leowyatt88 to add them if they aren't already.
@jpellis2008 @BorrisInABox @alexhall
Interesting, maybe we can get @Leowyatt88 to add them if they aren't already.
@jpellis2008 @BorrisInABox @alexhall
@FreakyFwoof @Scott @jpellis2008 @BorrisInABox @alexhall I will if they're not private. Some of the models on mvsep are not public.
@FreakyFwoof @Scott @Leowyatt88 @jpellis2008 @BorrisInABox @alexhall I’ve always wanted to know how people train these things.
I get that you have to put a tonne of Music through to basically teach it but how do you tell during training what you want the model to separate?
I would assume the models would have to be exported with all of those samples in there?
It's just always fascinated me.
I get that you have to put a tonne of Music through to basically teach it but how do you tell during training what you want the model to separate?
I would assume the models would have to be exported with all of those samples in there?
It's just always fascinated me.
@FreakyFwoof I don't think those ones are open source. @Leowyatt88 @jpellis2008 @BorrisInABox @alexhall