Traktor Pro 4 is here!

Exactly how I feel aswell. @Friedemann_NI

Hello, this is my first message and I registered exactly for this reason.

I just bought the Kontrol MK3 with the Traktor license a couple of days ago, I really need to pay again for the Traktor 4? does not make me happy to be fair.

Great respect for that new release and huge props for that new UI - it looks amazing!

I'm incredibly happy that this has been released, and I understand that as it's only version 4.0 (the first release), the software might not be absolutely perfect yet.

Especially impressive is that you managed to get the stem separation with all its AI models into such a small package size, but its still taking a lot of processing time.

Please forward that following hint to the developers to easily reduce processing time:
By compiling demucs with the cuda version of the torchaudio/PyTorch library, Demucs allows the use of CUDA processors. With CUDA, the separation time for me is around 2-3 seconds. If CUDA is not available, the CPU version is automatically used, which eliminates the task of differencing if the computer is capable of GPU processing.
Infos here:

There are also PyTorch libraries für Metal integration on the Apple Silicon series (via MPS), which are very efficient and can also be used.

I can share my package.json for that if needed.

Since CPU generation requires a lot of memory, its on both acceleration modes possible to work with segmenting.
More infos here:

It shouldnt be a big task since that project only need to be compiled with different but similar working dependencies.

If youre using Spleeter their are similar approaches aswell:

In general most of that stem seperation libraries work with similar dependencies. (tensorflow, numpy, torch..) That means that hardware acceleration is easy to implement into any stem seperation library which is python based.