Researchers at Pennsylvania State University have uncovered an unexpected way to tap into phone conversations—by picking up the subtle vibrations of a handset’s earpiece. Using a millimetre‐wave radar sensor, similar to those found in autonomous vehicles and modern 5G networks, the team managed to transcribe dialogue from as far as 3 metres away, raising a number of privacy concerns.
Although current transcription accuracy is only about 60%, the implications are serious. Doctoral candidate Suryoday Basak, the lead author, noted that these barely noticeable vibrations can reveal far more than we might think. The research aims to caution consumers about future vulnerabilities where such technology could be misused.
To transform these vibrations into text, the researchers adapted an AI speech recognition model, Whisper, originally designed to work with clear audio. By retraining just 1% of its parameters, they helped it better decipher the noisier radar data—a process not unlike how we’d use context to understand a muffled conversation.
If you’ve ever worried about privacy during a sensitive call, this research might prompt you to rethink the security of everyday communication tools.
