I've seen Veritasium's video that describes how sound can be picked up by a camera by observing small movements of objects sensitive to pressure changes (bag of chips, aluminium foil, paper...).
That got me wondering, how loud does a sound need to be to make pressure waves detectable by a camera ?
The example above is extreme because waves are adding up due to the bullet being supersonic, but it shows that a pressure wave with a high enough amplitude can be detected by a camera. So let's set up our virtual experiment :
Finding pressure waves in an image isn't obvious to humans, but suppose we have advanced image processing techniques that eliminates that factor.
Our camera, perfectly static, is looking for apparent deformations of black and white grid.
Suppose our camera has a frame-rate sufficiently high so it doesn't affect the experiment.
At what point will our setup see the sound ? Does the quality of the audio only depend on the quality of the camera ? If so, assuming camera technologies always improve, at some point, robots will be able to see people speak only by looking thew the air ? Are any of my assumptions broken ?
If I had to ask a single question, what are the limitations of this technique, if any.