Has he released his source code for that yet though? From what I've heard it's still not available anywhere.

As for voice recognition; it's pretty good but not nearly perfect yet. When you can limit the vocabulary though (as you can in most programming contexts) it does already perform rather well.