I actually used to work for a guy who was one of the pioneers of speech recognition biometrics. There are a LOT of issues with them (accents for one).
But it basically comes down to a sliding scale. If you want a piece of software to recognise EVERYONE's speech then it will not be very accurate.
If you can develop the software to recognise the speech of a SINGLE person, then an engineer can make it reasonably accurate.
I am sure it will get there in the end, but I forsee some problems with replacing keyboards with speech.
The fact that I am typing this post from a public library would be one problem.
