search

clovaai / voxceleb_trainer

In defence of metric learning for speaker recognition
MIT License
1.02k stars 272 forks source link

question about data transpose in SpeakerNet.py #92

Closed forwiat closed 3 years ago

forwiat commented 3 years ago

Hi @joonson , Thank you so much for the great work you shared. I am confused that why do "transpose(1, 0)" before feedward model in training. data = data.transpose(1,0). This code is in line95 SpeakerNet.py. Does it means transpose batch and frames?

joonson commented 3 years ago

Its basically transposing N and M in the paper, not frames.