Traktor stems separation is slow

Snorki

Hi, I have a question about splitting the track into stems. For a track that is 6 minutes, he needs 4 minutes to divide it into stems, which is not convenient to do during a mix. I tried with serato to add an unanalyzed track and boy was it ready to play the stems. why is the traкtor so slow?

lord-carlos

why is the traкtor so slow?

Because traktor uses CPU only, while all other software use the GPU.

🤷

I would not use it while DJing, only for prep of certain songs.

Stevan

Did you try tilting the laptop to the right? jk

I consider it as an offline process at the moment. Positive that it will be improved in future versions.

EnryDalo

https://community.native-instruments.com/discussion/33868/traktor-stems-separation-is-slow

Definitely is too slow! Native Instruments, please do something!

Terrordisco

Real time stem sep doesn't sound as good. I've used 'em all, and they can get pretty janky. And offline stem sep does shield older DJ laptops from obsolescence.

So offline stems are good. I've been using RipX to make offline stems for a year or two, to play in Traktor.

In other words, this isn't much of a feature.

But real time stem sep that sounds good would only work with pretty expensive computers, so…

Anyway, in the mean time, you need to do the offline stem separation when you're prepping your sets. Sorry.

Kubrak

And what about quality? Have you compared STEM separation quality of Traktor and Serato?

I do not say one or other is better or worse, I do not have experience. But my guess is Traktor might have better separation quality.

lord-carlos

NuoStems aka demucs algorithm should be equal or close to traktor, and yet can utilize GPU.

Someone in another thread said that iZotope (What traktor implemted) is based on demucs as well. I can't verify that, but that would make sense why they sound so similar.

Btw, there is no real time stem yet. They all do it quickly on load, not live.

Kubrak

I haven't tried. But even if they use exactly the same algorithm big difference may be in the training of this algorithm. And possibly also in "right" selection of set of algorithm parameters for given track type.