Hacked Microsoft Kinect trained to recognize objects
Using OpenKinect as an interface, OpenCV to process the captured images, CMU Sphinx to recognize speech and Festvox to synthesize speech, one hacker was able to create a program that recognizes random real-world objects and communicates with a user in real time. He notes in the video's description that it's just a 'proof of concept', but one wonders how long it actually took him to cobble it together because the video was uploaded only a few days after OpenKinect's release.
The most exciting prospect though, in my opinion, is that Microsoft must surely be working on their own in-house tools (and games!) to better utilize Kinect's potential. The cocky work of lone cowboy coders is one thing, but next-generation interfaces from powerhouses like Microsoft are something else.