Hi,
I would like to know if there is any issue in the inference results if I don't determinize my decoding graph. When I say "issue", I mean pretty much WER (and recognition time, if possible).
To give you some context: in my application, when doing inference, I use a hand made grammar and when I try to apply determinizeStar(LG) it takes some hours.
I understand determinzation as an optimization step that, among other things, saves time when decoding. But if I don't determinize, should I expect any difference in the recognition result? I think I might experience some slower decoding but the final result won't change. Is it correct?
Thank you,
Eduardo