search

ruizhecao96 / CMGAN

Conformer-based Metric GAN for speech enhancement
MIT License
309 stars 60 forks source link

ComplexDecoder #13

Closed lth456321 closed 1 year ago

lth456321 commented 1 year ago

Hello, I found that compared with your paper, your code has a conv layer reduced in ComplexDecoder. Could you tell me the reason or if I misunderstand your paper/code? Thank you.

SherifAbdulatif commented 1 year ago

Thank you for your note. We originally planned it as the paper to have 2 layers at the end one with 64 channels and then we shrink it to 2 by last layer. According to the trials, the last convolution layer is enough, I don't think having this extra 64 layer will affect the scores that much though.