FastSpeech2
FastSpeech2 copied to clipboard
Why is validation loss not decreasing after certain steps
I am wondering is there any particular reason for validation loss not to decrease after certain steps, Can you help me answering this question
same thing happen with me do you know the answer ? thanks
@zaynabmu @jayachandrakalakutagar hava you known why?I have the same problem,and when i use frame level features of pitch and energy,the qulity of synthesized audio is bad in inference phase.But the qulity of synthesized audio is good in trainning phase.