A Vibration Signal Enhancement Scheme for mmWave-Based Sound Eavesdropping

The emergence of millimeter wave (mmWave) sensing in sound eavesdropping poses a significant threat to speech privacy. It exploits the fact that speech can induce vibrations in objects, which in turn changes the phase of the reflected mmWave signal. Through extracting the phase change, the attacker is able to rebuild the speech. Conventional methods of phase change extraction, mainly based on circle fitting, often fail to consider the phase distortion caused by speech, which decreases the quality of reconstructed speech. To fill this gap, this paper proposes an innovative mmWave-based sound eavesdropping framework. The novelty lies in a vibration signal enhancement scheme to calibrate the phase distortion, which consists of the speech activity detection and the constrained circle fitting. In particular, speech activity detection divides the signal into speech segment and non-speech segment. Subsequently, the constrained circle fitting utilizes the fitted circle radius of the non-speech segment to determine the circle center of the speech segment, enabling precise phase change extraction. Our experimental results reveal that our proposed framework is able to recover speech with a similarity of 0.44 to the original speech and boost the signal-to-noise ratio (SNR) for sound-induced vibration signal by 13%.