xcmyz/FastSpeech

training stops in few seconds and no checkpoint file created

Ahmad-noborders opened this issue · 3 comments

Ahmad-noborders commented

hello @xcmyz, i am training on custom data and followed all these steps

  • Download and extract LJSpeech dataset.

  • Put LJSpeech dataset in data.

  • Unzip alignments.zip.

  • Put Nvidia pretrained waveglow model in the waveglow/pretrained_model;

  • Run python preprocess.py.

  • * if you want to calculate alignment, don't unzip alignments.zip and put Nvidia pretrained Tacotron2 model in the Tacotron2/pretrained_model

preprocess.py creates train.txt file and also creates files in mels folder
but train.py stops in only 5 to 6 seconds with following outpout

Model Has Been Defined
Number of FastSpeech Parameters: 51380417
Defined Optimizer and Loss Function.

---Start New Training---

and after that no checkpoint file created model_new folder is empty. any help is highly appreciated
thanks

xcmyz commented

try "unzip alignments.zip"

Ahmad-noborders commented

@xcmyz as i mentioned i have already tried with unzip alignments. but still same thing happens

xcmyz commented

I didn't get enough information to locate the problem. You need check your dataloader.