search

HarryVolek / PyTorch_Speaker_Verification

PyTorch implementation of "Generalized End-to-End Loss for Speaker Verification" by Wan, Li et al.
BSD 3-Clause "New" or "Revised" License
576 stars 164 forks source link

poor performance #46

Closed nidhal1231 closed 5 years ago

nidhal1231 commented 5 years ago

I am training the network on VCTK corpus (framerate =48kHZ 109 speakers with an average of 300 utterances per speaker) I got a very high EER ( 0.45) and I couldn't understand why the performance is poor (because of frame rate or there is not enough data or there is a problem in the model ) Any thoughts (for data augmentation I don't think adding noise because the model removes the noise in the data preprocessing ) Your help is much appreciated. Thank you.

nidhal1231 commented 5 years ago

@HarryVolek ,@mazzzystar @seandickert @xinli94 @Eason6wang, @fanlu @hujinsen , Any help please? Thank you