DeMixor - Separate Vocal From Any Song
I wanted to do this for a long time, ever since I started to understand machine learning. Here's the task: given a song with vocal and instrumental components, “demix” it so that you get separate tracks. Sounds simple, right?
Not so much. Not if you want to do a perfect job. Yeah, sure you can fire up Audacity and click some buttons here, adjust some knobs there. But you won't be happy with the result, most of the time. There are also professional solutions but they are not great either. Turns out, it’s still not a completely solved problem in 2020.
But we’ve made some damn good progress.
Machine learning based algorithms are pretty good at it now, although artifacts can still be heard. The quality also largely depends on the genre. That being said, the results provided should be good enough for most use cases.
But anyway, I'm happy to announce, after reading countless research papers, collecting and cleaning up an insane amount of data, doing expensive experiments with lots of computer time, and most importantly, sinking almost all my free time into this thing, I finally have a decent solution to this problem. Heck, you can even try it for free! Head here.