Comments

Until speech recognition employs lip-reading, speaker recognition, and facial recognition (all things even the human brain does automatically), as well as some more flexible recognition patterns (so we can speak more naturally), and perhaps a widened scope of understanding (using context clues from the entire conversation rather than just the last sentence), speech recognition will not be as natural as it should. If we really want the computer to respond like a human, it needs to know who's talking and if you're talking to IT (i.e. by name or if you're looking at it; camera). It also needs to be able to make educated guesses about what you're saying based on as many relevant context clues as possible. And lastly, when it tries to hear what you're saying, it needs as many sources of information as possible--hence the lip-reading. The McGirk effect should not take precedence over accuracy.