Side Eye gets audio from still images and silent videos

https://techxplore.com/news/2023-09-audio-images-silent-videos.html

"can determine gender and exact words of someone speaking in room where photo was taken even if they were off-camera... voice near camera lens causes tiny vibrations bending light... smart phone photographic method makes it easier, scanning 1 row of pixels at a time hundreds of thousands times in single photo. So amplifies > thousand times amount of frequency information, audio granularity... As long as there is even a little bit of light, even photo pointed at ceiling... trained with machine learning... cybersecurity"

Comment by Simon Derricutt at Revolution-Green:

It's helped along by the fact that Apple phones no longer actually take photographs as a single image, but as a short video sequence of multiple photos. Allows you to select the frame that's best, for example where people aren't blinking. If you read the description carefully, though, the claim of "determine gender and exact words" isn't quite true, since the retrieved audio is somewhat muffled. It's using the complete phone as the microphone, after all, and looking for small fluctuations in the light input. Still pretty amazing, though.

http://disq.us/p/2vyquic

Related: Getting dynamic information from static snapshots

https://phys.org/news/2024-04-dynamic-static-snapshots.html