search

xiph / LPCNet

Efficient neural speech synthesis
BSD 3-Clause "New" or "Revised" License
1.14k stars 295 forks source link

I have another question about training，can you help me see it，thanks! #131

Closed zengbing1992 closed 3 years ago

zengbing1992 commented 4 years ago

The u8 and f32 files generated by training, where u8 is a pcm file, then my question is whether the pcm file must be in u8 format? My understanding is to use the first three saved data of the u8 file for training, and then use the u8 file Is the fourth data calculate loss to adjust the weight of the model? I don’t know if my understanding is correct?

yingfenging commented 4 years ago

Yes,your understanding is correct, the u8 file must be u8 format, because the loss function is sparse_categorical_crossentropy ,

zengbing1992 commented 4 years ago

@yingfenging thank you so much!

justforhome commented 4 years ago

@yingfenging Did you train the LPCNet using the Chinese dataset ? Could you please tell me your email address ?

yingfenging commented 3 years ago

Please leave your email， and I will connect you.

@yingfenging Did you train the LPCNet using the Chinese dataset ? Could you please tell me your email address ?

wangshiyao-1119 commented 3 years ago

Please leave your email， and I will connect you.

@yingfenging Did you train the LPCNet using the Chinese dataset ? Could you please tell me your email address ?

Hello, I also want to use the Chinese dataset to train, can I discuss it with you? This is my email address：1067701367@qq.com