What materials did you find were most difficult to perform this attack on? I am guessing material like wood or rubber does not vibrate enough to observe a pattern? Also wouldn’t you be able to prevent this attack by having audio generators that generate near random audio signals that mask the data the attacker is seeking?
Right, good question. Anything that is very rigid and heavy doesn't vibrate enough when we play sound near it. I would say it's more about the thickness and weight of the target object than the material itself. For example, a very thin piece of wood would work much better than a wooden table leg. In the paper we actually test against ten different objects that are likely to be within reach of the robot vacuum cleaner.
The difficulty with masking this attack is that you need audio playing at a comparable volume near to the legitimate speech sources, which might be pretty disruptive. In the case of background white noise generators, for legitimate audio playing at around 70 dB SPL, we don't lose much accuracy until the background white noise exceeds 75 dB SPL.
Very interesting. All the new phones have lidar built in that works up to 30 feet away. In theory with the technology you applied here, it would be possible for them to eavesdrop without a microphone.This is almost as unsettling as the cameras that can visualize WiFi reflecting off human skin through walls. Its a matter of time before a commercial model hits the market. https://youtu.be/fGZzNZnYIHo what a brave new world we are about to live in.
