The point is speech recognition is becoming more and more common. But I've found that the accuracy is still pretty bad. By "pretty bad", I mean not 99% accurate. If it's not at least that accurate, I feel that I might as well type it out. As a sidenote, I have the same issue with predictive keyboards.
Considering there are a very limited number of pronounceable sounds (phonemes) in English, shouldn't we be able to vastly improve speech recognition by asking the user to speak out every phoneme as set-up?