4.0 Article

Wavesdropper: Through-wallWord Detection of Human Speech via Commercial mmWave Devices

Publisher

ASSOC COMPUTING MACHINERY
DOI: 10.1145/3534592

Keywords

word detection; mmWave sensing; through walls

Ask authors/readers for more resources

Most existing eavesdropping attacks that rely on sound waves for speech retrieval are ineffective in scenarios protected by soundproof materials. This paper presents a novel method using portable and commercial off-the-shelf mmWave devices to compromise human speech in isolated rooms, along with a word detection system called Wavesdropper that utilizes a mmWave probe to sense throat vibrations and recover speech contents. Experimental results show the effectiveness and accuracy of Wavesdropper in speech recognition.
Most existing eavesdropping attacks leverage propagating sound waves for speech retrieval. However, soundproof materials are widely deployed in speech-sensitive scenes (e.g., a meeting room). In this paper, we reveal that human speech protected by an isolated room can be compromised by portable and commercial off-the-shelf mmWave devices. To achieve this goal, we develop Wavesdropper, a word detection system that utilizes a mmWave probe to sense the targeted speaker's throat vibration and recover speech contents in the obstructed condition. We proposed a CEEMD-based method to suppress dynamic clutters (e.g., human movements) in the room and a wavelet-based processing method to extract the delicate vocal vibration information from the hybrid signals. To recover speech contents from mmWave signals related to the vocal vibration, we designed a neural network to infer the speech contents. Moreover, we explored word detection on a conversation with multiple (two) probes and reveal that the adversary can detect words on multiple people simultaneously with only one mmWave device. We performed extensive experiments to evaluate the system performance with over 60,000 pronunciations. The experimental results indicate that Wavesdropper can achieve 91.3% accuracy for 57-word recognition on 23 volunteers.

Authors

I am an author on this paper
Click your name to claim this paper and add it to your profile.

Reviews

Primary Rating

4.0
Not enough ratings

Secondary Ratings

Novelty
-
Significance
-
Scientific rigor
-
Rate this paper

Recommended

No Data Available
No Data Available