I have try to run the recipe for Librispeech 960hrs train set, it works well.
But after reduced the data to Librispeech 100hrs, I found it not working and give WER > 100% with validation set.
So would like to know if there is any recipe for Librispeech 100hrs or other train set with similar amount of data as reference ?
I have try to run the recipe for Librispeech 960hrs train set, it works well. But after reduced the data to Librispeech 100hrs, I found it not working and give WER > 100% with validation set. So would like to know if there is any recipe for Librispeech 100hrs or other train set with similar amount of data as reference ?
Thanks.