I have been interested in this research for a long time - can you recover audio information from image-only video?
This work was originally done at MIT by Dr. Abe Davis and colleagues. Abe was kind enough to revisit an experiment for me using code he hadn't touched in years. I learned a lot - it's way more sophisticated than looking for motion in the video (because these sound motions can't actually be seen - they're much smaller than a pixel).
This video had a significant sponsor so it is an unpaid post. I know some of you wouldn't mind supporting a post like this but I think it's best to conserve your resources. Sponsorships can be great especially when they stay out of the creative and allow me to do things I couldn't do easily on my own, like fly to San Jose and buy a camera in a pinch to get a higher frame rate.