That would be cool!
I don't want to throw a wet blanket on this idea as there are always possibilities.
The Androids use Googles speech recognition service AFAIK. The voice processing is all sever side.
Voice independent recognition takes some serious processing power though that power is getting closer all the time. Voice dependant is a lot easier, but very tedious to initially train. I has been a long time since I have researched this so maybe someone else can step in and fill in the blanks for us
Take care.