Can we speed up speech_to_text_eval.py?

[twmht]

I found out the evaluation is pretty slow, only 4709 files.

the gpu usage is very low, it seems most of the time was done on cpu. num_workers is set to 8.

any idea?

[titu1994]

There's so many variables - what model are you using, how long are your audio files, what is you batch size etc. Can't debug without that