Sign In:     

Forum: Wishes and new features

Topic: Stems separation v2
Hi guys,

The addition of stems has been a big upgrade to VirtualDJ.
However, i found that there are two main defects with this feature. Do you guys know any plugin to correct this, or do you know how to submits those ideas to the devs ?


1 - Stems separation quality

Even with stems preparation, the stems separations quality is... OK. But it's definitely not good enough for professional usage. I often find myself doing the stem separation by myself, which is a lot more complicated.
I suggest that the proprietary algorithm for stems separation could be improved by more powerful one, such as BS-Roformer (open source / MIT, https://github.com/lucidrains/BS-RoFormer ) --> gives great vocals / instru separation. If it was used first in the VirtualDJ current algorithme, the Acapella from virtualDJ would be absolutely amazing, way better than the other softwares.
After this first separation , virtualDJ could use an ensemble from other top open source algorithms reported on mvsep.com/algorithms could improve a lot the current proprietary algorithm.
2 - Stems separation storage

Stems are stored in pretty large mp3 files. The problem with this type of file is that it uses a lot of storage, for a poor quality.
Maybe a simple conversion to .opus file format, using the open-source ffmpeg library, would make the stems way bettter

Bests regards
 
Posted Wed 11 Jun 25 @ 2:47 pm
It is already v2 :)
Are you still using v1?
 
He's probably suggesting that algorithm be used for a v3 update.
 
A step aside: I have a dream


Oh! If stem file format could become a standard, just like mp3, wav, flac... are, that would be sooooo great.
Having this, everyone could choose their preferred tool for stem calculation (yes: that makes sense only if stem are prepared in advance, I know), then use them in any software...
 
Native Instruments already tried that route...didn't really get anywhere.

(IMO) Part of the trouble with the thought is given where we are now (AI based stem tech) is - if you open up the pre-prep to any tool, then what competitive advantage does the DJ software have versus others in that area (and ultimately, that removes one reason to choose one DJ software vs the other)?
This is just one of the avenues where each player is trying to play their cards to claim a win. Streaming platforms and access to them is another area.

In an ideal world where we cared about achieving the best product (no matter what it's called, where it's made or who uses it) for everyone to use for the task as the main goal, openness of choices would make all the sense, but unfortunately that's not how the world works currently.

As an interim though, most stem pre-prep files used in DJ software are just containers for each component of the song - you could probably just write a script to package it up in the format that each vendor understands (VirtualDJ has the stem creator that does just that for VirtualDJ, format).

However, a standard format? Doesn't seem likely, unless the proposer has the money to push its marketing, distribution and adoption.
 
DJ VinylTouch wrote :
Native Instruments already tried that route...didn't really get anywhere.


It didn't get anywhere at that time because it relied on having the artists/labels release tunes in their component parts - which of course was not popular for at least 2 reasons.

Firstly, major artists/labels are just not keen on releasing their copyrighted material "in pieces" to potentially be used willy nilly for bootlegs etc.

Secondly, the time/effort/money it would take for someone to go through the back catalogues and bounce down every multitrack to four parts, if they weren't concerned about point #1 above. How much would DJs be prepared to pay for stems of top artists?

Now though, there's really no reason to not have a standard format - but as usual they all choose to do their own thing.

BTW I was looking at the new Steinberg Spectralayers 12, which not only separates music into vocals, drums, bass, guitar, piano, sax/brass and other, but can also unmix the drums into kick, snare, toms, hi-hat, ride and crash. Would you rather have a standard format or be able to separate even more parts?!
 
groovindj wrote :

It didn't get anywhere at that time because it relied on having the artists/labels release tunes in their component parts


Yep didn't really want to go into the details. Asking for a standard format (alone) is a lesser ask, if it's just format, but the question of why would a vendor (already with a way of doing things, having different number of stem parts, for e.g. and no clear incentive of return for switching), want to do that, and who determines the standard?
And as you pointed out, other non-dj vendors have focused on even more fine grained separation (more parts).

I think at this point, unless the clear benefits and returns can be proven to everyone with an implementation, eatablishing a standard would most likely not happen.
 
I know it's “a dream” and how idealistic it is.
Yet, standards came up (either de facto standards or collegial standards) and that's why all DJ software are able to play MP3, FLAC, blah blah... or display a JPG format logo and so.

What makes a software better than another in stem separation is not the format of the stem's audio, but the algorithm for stem calculation.
As long as there's no standard, that makes room for sync tools like Mixxo, Lexicon... and for HDD constructor because we need more space to store stems for VDJ, stems for Engine, blah blah... and pain in the *** for multi-software users.
 
What you are saying is true but remember

  • Multi-application use is not really an aim for DJ software
    In fact, in most vendor cases aim for the opposite (database format, etc) to keep you locked in
  • It's true that the format itself may not be a competitive factor, but remember that different players have different implementations from the inception of AI stem part usage

    Most obvious to this is different number of parts.

    I think VirtualDJ is one of the only players who separate Percussion and Drums into different parts), which woulf mean even with a preferred format, you'd still have to make diffetent preparations in the format to use it with different vendor implemetations.


The other formats mentioned had complete descriptions of format (representation + metadata + extensions) as their proposal, and had large, general-use adoption for music representation/playback way before DJ sofware was a thing, so the DJ software naturally used them as that's what people know.
When that isn't done with wholesale adoption, you end up with what we have here for AI stems - vendors doing their own thing. Just look at how long it took the C++ to arrive, what was done before that and even where things are now for a more prominent example.

Maybe a standard will come forward, but I highly doubt it, as there are too many points against it for the influential players (artists don't like it, just adds more work for DJ software makers that doesn't have to be undertaken - meeting up to agree on it and then going back and implementing it).
 
 
Please make the STEM FX more like RecordBox or have a easy and fast ability to change the Stem FX.
 
Pretty sure that it's no more easier to add Stem FX in Rekordbox than it is in VirtualDJ, it's just that you to know a bit of VDJ script. It's also much more flexible in terms of the controls as well in VirtualDJ.

However, could you describe what exactly you would want replicated? "Make VirtualDJ like Rekordbox" doesn't really mean anything to the devs.