That's unlikely. How do Chinese people write a sentence on the computer? Old systems would have you write the letters, eg.: "shi", then display a long list of all the possible characters with that sound (for all 4 tones). Newer systems use context to figure out the right characters for the whole sentence at once (probably using naive Bayes --- see Peter Norvig's spell checker), so that I can just type "Wo shi Jianada ren" and the output would be the correct 6 Chinese characters. I never need to specify the tones.

So I would assume that voice recognition could do even better by analyzing tonal information.