"Sensors Can Tap Into Mobile Vibrations to Eavesdrop Remotely, Researchers Find"

Penn State researchers have demonstrated that they could detect the vibrations of a cell phone's earpiece and decipher what the person on the other end of the call is saying with up to 83 percent accuracy using an off-the-shelf automotive radar sensor and a novel processing approach. According to Mahanth Gowda, assistant professor of computer science and engineering, and Suryoday Basak, doctoral candidate, the demonstration, which is available in the Proceedings of the 2022 Institute of Electrical and Electronics Engineers Symposium on Security and Privacy, reveals a significant security concern. Basak emphasizes that as technology becomes increasingly reliable and robust, adversaries will be more likely to exploit such sensing technologies. The radar works in the millimeter-wave (mmWave) spectrum, specifically in the bands 60 to 64 gigahertz and 77 to 81 gigahertz, which prompted the researchers to call their method "mmSpy." The researchers simulated people speaking through the earpiece of a smartphone in the mmSpy demonstration. According to Basak, the brand is unimportant, but they tested their approach on a Google Pixel 4a and a Samsung Galaxy S20. The phone's earpiece vibrates as a result of the speech, and the vibration spreads throughout the phone. They used the radar to detect the vibration and reconstruct what the person on the other end of the line said. Their method works even when the audio is inaudible to humans and nearby microphones. This is not the first time similar vulnerabilities or attack modalities have been discovered, but this aspect of detecting and reconstructing speech from the other end of a smartphone line has never been investigated, according to Basak. In order to remove hardware-related and artifact noise from the data, the radar sensor data is pre-processed using MATLAB and Python modules. The data is then fed into Machine Learning (ML) modules trained to classify speech and reconstruct audio. The processed speech is 83 percent accurate when the radar detects vibrations from a foot away. The accuracy decreases as the radar moves away from the phone, dropping to 43 percent at six feet. This article continues to discuss the demonstrated use of off-the-shelf sensors to remotely detect and identify speech on the other end of a cell phone call. 

PSU reports "Sensors Can Tap Into Mobile Vibrations to Eavesdrop Remotely, Researchers Find"