GPU RAM Issue about conv-tasnet

[SashankMLAI]

Also, wav files are being separated and giving results for models which I trained myself, but when i used your pretrained model and tried to separate sp1 and sp2, they are separated but the separated files are empty. No sound is produced.

[kaituoxu]

1. If your wav file is too long, the Conv-TasNet will be bigger, but GPU memory is limited, of course, so you will face out-of-memory(OOM) issue. To avoid OOM issue, you can try to use CPU to do validation.

2. The pretrained model is trained on wjs0 dataset. If you use it to separate music, of course you will fail.