ZFTurbo / MVSEP-MDX23-music-separation-model

Model for MDX23 music separation contest
564 stars 87 forks source link

Got this message when trying to run the command #37

Open DJTonydacoder opened 1 month ago

DJTonydacoder commented 1 month ago

I did everything I was told in the GitHub website buy it keeps saying "cannot run inference.py" Screenshot 2024-05-25 204009

I change "inference.py" to its file path but it still didn't work... Screenshot 2024-05-25 204651

Please can anybody give me a solution?

ZFTurbo commented 1 month ago

The path for inference.py is incorrect

ZFTurbo commented 1 month ago

You need to install requirements.txt or at least just numpy. pip install numpy

DJTonydacoder commented 1 month ago

You need to install requirements.txt or at least just numpy. pip install numpy

I got another problem, I tried processing 32-bit .wav files with overlap_large and overlap_small set to 1.0. Several hours passed but no results... I got this message, I don't remember exactly but it said it can't process 32-bit .wav files or something

ZFTurbo commented 1 month ago

Try with overlap 0.5 or 0.8. Code works normally with any wav files.

DJTonydacoder commented 1 month ago

But you said overlap 1.0 gives you better quality separation...

DJTonydacoder commented 1 month ago

Try with overlap 0.5 or 0.8. Code works normally with any wav files.

On your GitHub it says, overlap closer to 1.0 means better quality separation. I don't want to sacrifice quality for time but I still want the code to work

DJTonydacoder commented 1 month ago

Try with overlap 0.5 or 0.8. Code works normally with any wav files.

Alright, I tried again with low values for overlap this time, still took forever and no result. I also got this message, what does this mean? Screenshot 2024-05-27 173931

ZFTurbo commented 1 month ago

I see you use CPU it will be very slow. Try with smallest overlap possible like 0.1, to check if it finish. If you see it work you can try 0.99. But don't use 1.0 - it's like infinity.