Intel has an article on their site describing "Audio Visual Speech Recognition" and the technique employed to achieve higher reliability in recognizing human speech.

The kicker is that there is open source code available as well. Way to go.