Bren F
Is it possible to increase the quality of the stems 2.0 on a pre-processed basis? Perhaps a third engine for powerful computers with an option to export/save the stems as WAV for DAW editing. Essentially getting the quality up to the level of Logic pro.
Inviato Thu 02 Jul 26 @ 3:11 pm
IIDEEJAYII
+1
Inviato Fri 03 Jul 26 @ 6:22 am
djcel
You can already read the vdjstem file with audacity for example. Just drag'n'drop the file in audacity
Inviato Fri 03 Jul 26 @ 9:27 am
IIDEEJAYII
Thanks for the reply :)
Yes one can easily get the poorer stems (1.0) and the better stems (2.0).
The wish (or at least mine) is to have much better separation, maybe a 3.0 or 4.0 version, that would never be used live, and could take 30+ min, (instead of 30 sec w 2.0), for a few tracks that one would, love, say, have an amazing acapella, with fewer artifacts, and more clarity, which can be heard in the complete track, but lost in the 2.0 vocal stem, not fully capturing it all.
Its a quality issue, not an issue of finding the stems.
Yes one can easily get the poorer stems (1.0) and the better stems (2.0).
The wish (or at least mine) is to have much better separation, maybe a 3.0 or 4.0 version, that would never be used live, and could take 30+ min, (instead of 30 sec w 2.0), for a few tracks that one would, love, say, have an amazing acapella, with fewer artifacts, and more clarity, which can be heard in the complete track, but lost in the 2.0 vocal stem, not fully capturing it all.
Its a quality issue, not an issue of finding the stems.
Inviato Sun 05 Jul 26 @ 10:06 pm
DJ VinylTouch
Why not use the better tool to create audio stems (or better yet, get stems from the producer where you can) and then use the built-in stem creator tool to create a prepared stem file based on those?
The amount of work being asked to produce a better separating model might be quite large compared to the return on investment for the majority use case (much longer analysis time (making realtime use potentially unfeasible) + processing power for a very small selection of users).
Summary: There is potential for stem separation improvement, but there are consequences for that to be achieved, and VirtualDJ is meant to be a realtime DJ performance tool (there is a tradeoff here).
The amount of work being asked to produce a better separating model might be quite large compared to the return on investment for the majority use case (much longer analysis time (making realtime use potentially unfeasible) + processing power for a very small selection of users).
Summary: There is potential for stem separation improvement, but there are consequences for that to be achieved, and VirtualDJ is meant to be a realtime DJ performance tool (there is a tradeoff here).
Inviato Sun 05 Jul 26 @ 10:30 pm
IIDEEJAYII
Agree, 100% this is not a request, (as mentioned) for realtime processing, and would never be needed for typical mixing.
I would never want to wait 30+ min during a set to get a better separation that would not be noticed.
Asking about for a one-off separation, to create, say, a voice stem, which may not be available to purchase. The purpose is to create a special effect, or unique mix, one wishes to create, (wo the artifacts).
(there are acappellas for many popular tracks)
I assumed that the stem separation was tuned to create the best result given the time constraints of making something useful. And understand there are diminishing returns as the stems 2.0 are not that much different from the stems 1.0 considering, 2.0 takes 30+ times longer, and assume to get to the next step up would require at least a 60-100X of the 2.0.
(actually don't know the math, for the improvement, it could be quality scales with T (to second power) or higher )
Just asking for the similar algo to added without the time constraints, maybe.
As you suggest, maybe should explore other places/ways to create a better stem separation, for those not available.
I don't mind spending the time, electrons and GPU to make it happen, for the infrequent, and otherwise not available acappella.
I would never want to wait 30+ min during a set to get a better separation that would not be noticed.
Asking about for a one-off separation, to create, say, a voice stem, which may not be available to purchase. The purpose is to create a special effect, or unique mix, one wishes to create, (wo the artifacts).
(there are acappellas for many popular tracks)
I assumed that the stem separation was tuned to create the best result given the time constraints of making something useful. And understand there are diminishing returns as the stems 2.0 are not that much different from the stems 1.0 considering, 2.0 takes 30+ times longer, and assume to get to the next step up would require at least a 60-100X of the 2.0.
(actually don't know the math, for the improvement, it could be quality scales with T (to second power) or higher )
Just asking for the similar algo to added without the time constraints, maybe.
As you suggest, maybe should explore other places/ways to create a better stem separation, for those not available.
I don't mind spending the time, electrons and GPU to make it happen, for the infrequent, and otherwise not available acappella.
Inviato Mon 06 Jul 26 @ 6:37 am