Voice Command Recognition: State of the art?

I found (like 10 years ago) playing with the MS Speech SDK that individual command words worked fairly well (limited vocabulary) when just random speech recognition had something like a 25% failure rate (I repeated text heard on a TV show in their trasncription sample program and wasnt pleased with how many wrong words were processed).

I recall the CPU usage at that time was 200m instructions per second of the available CPU (was a single core)

If we ever want games with more command options, something like this will be needed (as well as issuing short sentence commands to lakeys instead of endless use of menu trees)

Working out input seperation between interplayer communications might not be horrible (a mic button ?)

I hope that the random text transliteration is also better now, as I miss open communications in MMORPGs (like UO, with text over the avatars head instead of the distraction/crudity of a chat box) and the bandwidth issue (seeing 100 conversations at once - and now getting rid of the limitation of only being able to play or talk but not both at the same time).

Thanks, @wodinoneeye.

The one thing that bothers me even more than a few missed words is the mic button. I really dislike the concept, though it's necessary for unobscured recognition and is, anyway, a staple in multiplayer gaming. One more hand / finger that has to be free to press one more button seems far from ideal... that's a big part of why I want to implement voice recognition, to avoid having to remember tons of keys and key combos.

Given that there's no other practical way (that I know of) for clear recognition, I'd rather make that sacrifice and not have to deal with menu trees which are a UI abomination... or at least do so only as a fallback measure.