I'm one, and I'm laughing hysterically. Speech reading depends on context + partially-heard speech sounds + (to get accurate) knowing the other's individual speech quirks + prediction + likelihood-testing. What you SEE is integrated into a whole that includes all of the above; then there's a bit of trained neurological synesthesia and out comes hearing. Sort-of. My wife and I have fun when I tell her what I think she mumbled and she cracks up at the garble. And here I'm all trained and so forth. I'll believe a machine can do it when it successfully "hears" a mumbler with an accent, foreign-language word order and misphrasing and a droopy mustache covering half his mouth.
Biting the hand that feeds IT
