I run a code in my laptop (CPU i7, about deep learning, speech recognition ASR and) the result(model train) is ok as I expected: 0-17: 0 is unknown words(fillers) and 1-16 are my keywords, 17th is blank and their present show that what percentage of recognition is correct.
then I run this code (same code ,same dataset) in my friend laptop(GPU RTX 3080) to speed up my work. the version of CUDA and other toolkits and libraries is compatible and in this sense there is no error or problem (I THINK). result is not ok and its looks that it cant read any signal and just predict blank(number 17)
DO you have any ideas about this problem? if yes, please tell me how can solve it, thank you.
I checked the code and run it with different epoch and increased the patience, but it didn't work.