search

YChenL / DS-TDNN

Official implement of "Dual-stream Time-Delay Neural Network with Dynamic Global Filter for Speaker Verification" in PyTorch
https://arxiv.org/pdf/2303.11020v2.pdf
39 stars 7 forks source link

how to run MFA-TDNN? #3

Open wcqy-ye opened 10 months ago

wcqy-ye commented 10 months ago

Hello, I'd like to ask how to run the MFA-TDNN model. I attempted to run it using the following approach, but encountered an error. Thank you for sharing your code, and I look forward to your response. image I believe the size of the input x meets the requirements for input to the MFA module. image

FiLafs commented 9 months ago 
    for i in range(self.scale):
        att_TDNNs.append(att_TDNN(width, D, **CE**))
       Change the parameters , they should be
   for i in range(self.scale):
        att_TDNNs.append(att_TDNN(width, D, **width**))
FiLafs commented 9 months ago

Additionally, the MFAmodule section in this code does not use any BN or activation, which may lead to gradient issues