A few days before I was scheduled to leave for Apple's WWDC in San Francisco my boss had this idea that we should really have a demo application to show around there. So a couple of friends and I spent 3 very long days and built this:
The app streams the audio request to our servers, we run a speech recognition system on it, query google with the text and send the results back to the phone. We were all a bit surprised how smooth the app turned out and how well the system works. You can literally say anything. Here is another video of my boss showing the demo (watch the bit at the end).
This all builds on some work we did last August (see Gizmodo) when we worked at a scrappy startup and used a jailbroken iPhone. This time we were all legit and used the official SDK.