Back in the ‘early’ days (2012) of video processing, before we had AI algorithms, we often just pursued straight video processing techniques to tease out interesting information.
Enter MIT’s work on Eulerian Video Magnification. The underlying concept is to take small per-pixel differences from frame to frame, then magnify the changes, to detect things not ordinarily visible to the human eye.
One of the most powerful effects is the ability to accurately detect the pulse of someone, or the vibration of objects by sound – with just normal video footage. Check it out in their original demonstration video below:
In 2014 they did a Ted talk that shows some further interesting applications – including re-generating the sound going on in a room from the vibrations of the objects in it.
So, since 2014, you can even recover sound from a discarded bag of chips behind soundproof glass.