Speech recognition is fraught with problems, he says, such as recognizing variations in accent. Translation, too, is problematic, since accuracy depends on context and dealing with the peculiarities of each language.
“It’s clearly making mistakes, and sometimes the output is awkward, but that’s not the point. If you imagine being in a lecture and understanding nothing, then a little awkwardness is worth it,”
said Alex Waibel.
Leave a Reply