[Cryptography] "The Visual Microphone: Passive Recovery of Sound from Video"

[Jerry Leichter]

On Aug 5, 2014, at 4:35 PM, Christian Huitema <huitema at huitema.net> wrote:
> Also, even if the signal is hard to understand by humans, it might be
> possible to use machine learning and train a recognizer directly on the
> output signal...
Stories about listening to conversations in a room by bouncing a laser of the window have made the rounds for years.  Someone - maybe MythBusters - set out to test whether this actually worked.  And they found it didn't - all they got was noise.

But as we know from this bit of work, they were missing something fundamental:  Just because the unaided human ear - or even the human eye, looking at an oscilloscope - can't pick out the signal doesn't mean it isn't there.  Had they done the theoretical work and the signal processing it pointed to, they would undoubtedly have reached the conclusion that the "laser and window" microphone is a great way to listen in on a room.  After all, the signal was undoubtedly *much* better than even the high-speed camera image - much less the low-speed, edge-of-the-scan image.
                                                        -- Jerry