The idea is to make a proof-of-concept device that shows that you can obtain an audio signal from a video feed. This would be useful for eavesdropping (assuming you didn't have a mic for some reason), or for getting audio using a telephoto lens at long distance - farther than the sound would travel. A low-mass, thin object like a piece of paper, metal foil, or plastic could change an audio signal into an observable signal that could be captured in video and converted to audio. A known technique for eavesdropping is to bounce a laser off a glass window: the audio signal causes the window to vibrate, modulating the reflected optical signal.
http://scitoys.com/scitoys/scitoys/computers/laser/free_space_laser_data_transmitter.html
Finally found the link that covered exactly that idea http://newsoffice.mit.edu/2014/algorithm-recovers-speech-from-vibrations-0804 I would even say this was on the hackaday.com blog once. Ha, found it, now that I know what to look for :) http://hackaday.com/2014/08/06/focus-your-ears-with-the-visual-microphone/
Would be cool to see that redone here. They even got recognizable sounds from 60fps, which is quite remarkable, given the low sample rate for sound. Raspberry PI cam is able to film at 90fps per second. I won't say you'll be able to see a live stream, but may be it's possible to process the informations in an acceptable amount of time.