CRII: SaTC: Fingerprinting Encrypted Voice Traffic on Smart Speakers

University Of Cincinnati Main Campus, Cincinnati OH

Investigators

Abstract

Millions of users interact with smart speakers every day. However, there remains a significant gap in the understanding of the privacy impacts of smart speakers. A poor understanding of the privacy impacts can lead to unauthorized disclosure, affect the well-beings of users, and thwart Internet freedom. To bridge this gap, this project investigates the privacy leakage of smart speakers under a new encrypted traffic analysis attack, referred to as voice command fingerprinting, and develops new defenses against this attack. This attack infers which voice command a user says to a smart speaker by analyzing side-channel information of encrypted network traffic. This research includes four thrusts: (1) producing large-scale datasets for encrypted traffic analysis on smart speakers; (2) leveraging deep learning in the attack to investigate the privacy leakage; (3) promoting the efficiency of defenses by analyzing which encrypted packets should be protected with a higher priority; (4) developing a defense against the attack by generating adversarial examples on the fly. The research will promote the understanding of the privacy impacts of smart speakers and advance the knowledge in privacy-preserving technologies. The research findings will be disseminated through publications and presentations. The datasets and source code will be made publicly available for the research community. This project will integrate the research activities into curriculum development, render research opportunities to female and underrepresented students, and advance research experience for high school teachers and students in STEM (Science, Technology, Engineering and Math). This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.