I'm assuming it would need to be extremely high frame rate video to extreact decipherable voice data. Because even at 60 frames per second, which I understand is at the high end for normal video, that gives you a maximum frequency of only 30Hz, which is nearing the low end of human perception.

So it's interesting and a definite risk if an attacker is supplying their own camera, but definitely does NOT mean you can pull voice data from the vibrating tablecloth in a YouTube video, right?