I just came across this article by Jon Arking which does a pretty good job of describing what's in the Speech Application SDK Beta 3, and there is some good discussion about how Speech might be able to be incorporated into existing solutions (and why).