Researchers create AI-powered radar that listens to phone conversations from a distance

Researchers at Pennsylvania State University have developed a new technology that uses millimeter-wave radar and artificial intelligence to listen in on phone conversations from up to 10 feet away. The system recognizes and transcribes conversations with about 60 percent accuracy by analyzing the very faint vibrations that the phone's receiver makes during a conversation, Interesting Engineering reports.

The research is based on an adaptation of the open-source Whisper speech recognition model, which was retrained on radar data, which allowed to improve the quality of speech recognition, even though this data is very noisy and of low quality. The radar was installed approximately three meters from the phone, and the resulting data was processed by artificial intelligence, which could recognize up to 10,000 words.

While the system's accuracy isn't perfect, even partial recognition of keywords could have serious privacy implications. The researchers compare their technology to lip reading—it also doesn't recognize all the words, but it can help you understand the gist of a conversation, especially when you consider the context.

Importantly, this technology raises concerns about potential privacy threats, as attackers could use it to remotely listen to private conversations without the user's knowledge. The researchers themselves emphasize that their work aims to warn society about such risks and contribute to the development of protections against this type of espionage.

The research was supported by the US National Science Foundation, and the team plans to continue working on protecting personal information from the latest technological threats.