Researchers from Pennsylvania State University have created a new technology that employs millimeter-wave radar and artificial intelligence to eavesdrop on phone conversations from a distance of up to three meters. The system can recognize and transcribe conversations with an accuracy of about 60% by analyzing faint vibrations produced by the phone's receiver during calls. This is reported by Interesting Engineering.

The foundation of this research is the adaptation of the open-source speech recognition model Whisper, which was retrained on radar data to enhance speech recognition quality despite the noisy and low-quality nature of the data. The radar was positioned approximately three meters from the phone, and the data was processed by artificial intelligence capable of recognizing up to 10,000 words.

While the system's accuracy is not perfect, even partial recognition of key words can pose serious privacy risks. Researchers compare their technology to lip reading — although it does not recognize every word, it helps grasp the essence of the conversation when context is considered.

Importantly, this technology raises concerns regarding potential privacy threats, as malicious actors could exploit it for remote eavesdropping on personal conversations without users' knowledge. The researchers emphasize that their work aims to raise public awareness about such risks and contribute to the development of protective measures against this kind of spying.

The study is supported by the National Science Foundation of the USA, and the team plans to continue working on safeguarding personal information from emerging technological threats.