CLI slower than GUI

[mediarl]

Hi,

Thanks for the script. It's working fine except it's slower using the CLI than the GUI.
For the same song, it takes 22 seconds using the GUI and 54 seconds for the CLI.
I guess it's related to the options.
On the GUI, I only changed those options, every other is selected by default:

• Process method: MDX-Net
• MDX-NET Model: UVR-MDX-NET Inst HQ 1
• GPU Conversion

Any idea how to improve the speed on CLI ?

[mediarl]

I noticed it was using the other model (HP-UVR) and changed it for model_path='uvr5_weights/UVR-MDX-NET-Inst_HQ_1.onnx'
But it seems that for using an ONNX model vs a Pytorch one, it would require some changes.

[seanghay]

@mediarl I have another project for MDX-Net. https://github.com/seanghay/uvr-mdx-infer/blob/30d837ff95115ce25bf7059b8296bbf37febf7f1/separate.py#L78C26-L78C42

Right now it only uses CPU and if you want it to use GPU you can specify providers=["CUDAExecutionProvider"] and install pip install onnxruntime-gpu