search

tuanh123789 / AdaSpeech

An implementation of Microsoft's "AdaSpeech: Adaptive Text to Speech for Custom Voice"
95 stars 27 forks source link

nan loss during training #5

Open dominicmini opened 2 years ago

dominicmini commented 2 years ago

hi you, I'm trying to train model AdaSpeech from your project. However when I train pretrained model I get nan loss during training. do you have any way to fix it from your code? image

tuanh123789 commented 2 years ago

Yes, i'll update the newest code version to fix it soon. Because the new version of MFA tool make the preprocess code generate 0 frame for some phoneme.

tuanh123789 commented 2 years ago

in preprocessor/preprocess.py replace function average_mel_by_duration by this code: image

dominicmini commented 2 years ago

Thanks a lot, I fixed it! Do you have a pretrained model training on English or other language?

manhph2211 commented 2 years ago

@dominicmini I met the same issue but the above solution does not solve the problem 😢 Do you have any advice for me ❤️

manhph2211 commented 2 years ago

@tuanh123789 Hi bro:v Can you tell me about the effectiveness of the average_mel_by_duration function 💯

HandsLing commented 1 year ago

@tuanh123789 Hi bro:v Can you tell me about the effectiveness of the average_mel_by_duration function 💯

i met the same issue, do you solve this question?

tuntun990606 commented 1 year ago

@tuanh123789 Hi bro:v Can you tell me about the effectiveness of the average_mel_by_duration function 💯

i met the same issue, do you solve this question?

Hello, I encountered such a problem at the beginning, and now it has been solved. At the beginning, I made changes completely according to the pictures replied by the author. The preprocess.py 338 line still reported an error. Later, I deleted the last comma of the author, so preprocess.py would not report an error. Then you can train normally