search

wenet-e2e / wesep

Target Speaker Extraction Toolkit
101 stars 12 forks source link

some problem occured when train #4

Closed wendongj closed 1 month ago

wendongj commented 1 month ago

Dear, author, thanks for your great work. when I try to train the model, it have many warning like this: error to parse id10536/aZT_gTu6ilw/00005.wav.wav. I check the voxceleb dataset seems ok, did you have encounter such problem before?

I check the wav.scp file seems ok, it like follows: id10001/1zcIwhmdeo4/00001.wav /home/jwd/code/dataset/voxceleb/VoxCeleb1/wav/id10001/1zcIwhmdeo4/00001.wav

mrjunjieli commented 1 month ago

Hi, Maybe you could try to use torch version<2.0. In my case it works.

wendongj commented 1 month ago

Hi, Maybe you could try to use torch version<2.0. In my case it works.

I see, thanks, I will try it

wendongj commented 1 month ago

Hi, Maybe you could try to use torch version<2.0. In my case it works.

Hi, I can train now, and I use voxceleb1 to train, when I modify the reverb_prob in config file to 1.0, the other is not changed, the model is hard to convergence, and the loss is around -0.12, did you encounter this problem?