christianwheel
Now that Serato has added stems, I'd like to see VDJ keep its lead by offering higher quality stem extraction. I'd like to encourage the devs to look at the DEMUCS algorithm over Spleeter, as it provides a significantly cleaner output. I've implemented both algorithms in my own software and it is fairly similar to implement as Spleeter. It would be nice to keep the existing competitive edge over Serato.
Posted Tue 11 Oct 22 @ 10:40 am
klausmogensen
VDJ is not using Spleeter
It has it's own one created by Atomix (or at least modified by Atomix)
It has it's own one created by Atomix (or at least modified by Atomix)
Posted Tue 11 Oct 22 @ 10:56 am
christianwheel
Fair enough, in any case, the beta of Serato does seem to have cleaner sounding stems than VDJ, as a whole. Obviously it'll vary from song-to-song, I just think it would be prudent to allocate some development resources towards improving the separation is all.
I've written my own software using Demucs to create multichannel .vdjstems files that I can use with VirtualDJ on the songs where the built-in separation doesn't do too great a job. It's a lot of work and takes a lot of hard drive space.
Stems are amazing, don't get me wrong. I'm certainly not taking them for granted, nor speaking ill of the feature in any way. But there have been times when I've been playing and I find myself thinking, "wow, that sounds bad" when I'm isolating vocals. Certainly not expecting miracles. But algorithms exist (some even with a very permissive MIT license) that can sometimes do a better job, albeit at the cost of extra cpu processing. But many of us (especially those who do this professionally) have CPU cycles to burn. My M1 Max doesn't even break a sweat mixing 1080p video all night.
I've written my own software using Demucs to create multichannel .vdjstems files that I can use with VirtualDJ on the songs where the built-in separation doesn't do too great a job. It's a lot of work and takes a lot of hard drive space.
Stems are amazing, don't get me wrong. I'm certainly not taking them for granted, nor speaking ill of the feature in any way. But there have been times when I've been playing and I find myself thinking, "wow, that sounds bad" when I'm isolating vocals. Certainly not expecting miracles. But algorithms exist (some even with a very permissive MIT license) that can sometimes do a better job, albeit at the cost of extra cpu processing. But many of us (especially those who do this professionally) have CPU cycles to burn. My M1 Max doesn't even break a sweat mixing 1080p video all night.
Posted Wed 12 Oct 22 @ 4:48 am
klausmogensen
christianwheel wrote :
I've written my own software using Demucs to create multichannel .vdjstems files that I can use with VirtualDJ on the songs where the built-in separation doesn't do too great a job. It's a lot of work and takes a lot of hard drive space.
I've written my own software using Demucs to create multichannel .vdjstems files that I can use with VirtualDJ on the songs where the built-in separation doesn't do too great a job. It's a lot of work and takes a lot of hard drive space.
You can actually do this directly in VDJ now:
Posted Wed 12 Oct 22 @ 7:09 am