EMTap: Eavesdropping on Sound Transmitted via Audio Cables by Sensing Electromagnetic Variation Using RFID

When sound signals are transmitted via audio cables in form of current signals, detectable electromagnetic leakage is emitted around the cables, which could reveal the sound content being transmitted. This paper presents a sound eavesdropping system, EMTap, which senses electromagnetic variation around audio cables leveraging an RFID tag attached to an audio cable for sound content recovery. In some audiovisual scenarios with audio cables, such as offices and meeting rooms, it is assumed that attackers secretly attach a small, battery-free RFID tag to an audio cable without being noticed. Meanwhile, RFID readers are camouflaged as decorations/public facilities placed in/out of rooms to transmit and receive RF signals. When the sound signals are transmitted via audio cables, EMTap first activates the RFID tag to capture the electromagnetic variation patterns around audio cables upon RF signals. Then, EMTap reconstructs sound spectrograms from RF signal-based electromagnetic variation patterns utilizing a designed cross-modal Generative model, Nonlinear Stable Diffusion Model (NSDM). Finally, EMTap converts the sound spectrograms to sound content through sound recognition API. Extensive experiments demonstrate that EMTap achieves an average Mel-Cepstral Distortion (MCD) of 5.49 and Word Error Rate (WER) of 16.19% for eavesdropping on sound transmitted via audio cables.