search

TaoRuijie / ECAPA-TDNN

Unofficial reimplementation of ECAPA-TDNN for speaker recognition (EER=0.86 for Vox1_O when train only in Vox2)
MIT License
594 stars 113 forks source link

about using pretrain model to continue training on the other data #56

Closed phchtl closed 1 year ago

phchtl commented 1 year ago

i want to ask should i change the n_class（the number of speaker）， if i only change the n_class，it will go to error because of the difference of embedding size when loading the pretrain model.Thank you so much!

TaoRuijie commented 1 year ago

If you change the n_class, the classification layer is different. You should load the pretrain model without the classification layer only.