split to stems started but failed exit code 1

[michaelfullmer]

anyone know what this means?

[pough]

IIRC it has something to do with the resulting stem files not saving. I can't remember what, if anything, solved the problem. I think for me it just started to work after a number of tries.

The last few pages of this thread discuss it:
viewtopic.php?t=597455

[mhelin]

Problem with "Split to Stems" is that the extension tries to use the python environment stored in Windows
"C:\Users\Public\Documents\Tracktion\Waveform\Content\Extensions\Stem Separation\windows\python" but doesn't take into account the the user might have another Python version installed on the same PC. I guess it could be fixed using virtual Python environment(s). I'd like to use the Python version installed on my PC because it is set up to use Pytorch with the GPU (RTX 5060 Ti in my case) support.

[mhelin]

Btw. the stem split uses htdemucs, that's Demucs4 HT (vocals, drums, bass, other).
See https://github.com/facebookresearch/dem ... /ht/demucs

[Watchful]

You should open a trouble ticket with Tracktion; there's no one here who can help you, but we do appreciate the warning.

[mikoatkvr]

I have a hunch that it fails because you don't have enough RAM left at that particular time. Maybe a lot of cached files can be the cause. Try with a really small snippet of audio to check ( make sure that you have consolidated an audio 1st if you have just used the clip's edge markers for resizing , as the whole file is still processed ).

Also use you system monitor during the seperation to check what resources are being used.

[mhelin]

Opened a ticket (or two actually), so far no responses. I have plenty 32Gb of RAM btw. However, I was able to run the stem split from the command line (there is a shell script demucs_tracktion.bat in the Windows extension installation which I think Waveform runs internally) after setting up a virtual environment (using python -m venv <dir>) and PYTHONPATH and PYTHONHOME. The stem separation runs quite slow without GPU support, the original Facebook implementation from the github linked above cannot use the GPU, but I've been also using UVR5 (Ultimate Vocal Remover from https://github.com/Anjok07/ultimatevocalremovergui) which supports also GPU with the htdemucs 4s. I guess Cubase Pro / Steinberg and Studio One / Presonus use the same implementation under the hood (or bonnet in UK, long live Wheeler Dealers).

[talby]

I can reproduce this issue. STEM separation fails with exit code = 1. But this failure only occurs for a fresh installation of this extension package. For users having instaled this extension package in the past, this issue does not become visible.

I also openend a ticket about it.