So this is a thing:
A team of researchers from Adobe, Microsoft and MIT have developed a visual microphone algorithm that picks up audio by looking for microscopic vibrations in video footage. The technique exploits the rolling shutter effect in digital cameras (where the sensor reads pixels one row at a time) to detect sound-related movements that might otherwise be invisible; the only gear you need is a camera that can record at high frame rates. It’s good enough to capture singing from a bag of potato chips, and musical tones from a potted plant.
The first generation of this technology has trouble making out actual words, although it can identify voices. I would suggest we nuke Redmond and MIT from orbit before the second generation can be developed, just to be sure.