Spleeter 4 stem model
Hey! First off, amazing work. I was able to get a version of the 2 stem ml model working.
I'm wondering if you ever got a 4 stem version of the pytorch implementation working? I tried using the 4 stem model with 4 instruments, but it just produces garbage.
Did you ever get this working? Do you have any pointers on how to get this working?
Hello, thatcherclough. 4stems can be converted into a Core ML model and work correctly;
I have already tried successfully. Although there might be a few minor flaws, you can still distinguish the drums, bass, piano, and vocals.
To convert 4stems, you should refer to spleeter 4stems config to update your up_block/down_block in unet.py , note that 4stems use nn.ELU conv_activation