Nate Mathews

Nate Mathews

PhD Candidate specializing in the application of machine learning and deep learning techniques to address cybersecurity challenges, particularly in network traffic analysis.

WhisperVoiceTrace: A Comprehensive Analysis of Voice Command Fingerprinting

Published in ACM Asia Conference on Computer and Communications Security (ASIA CCS), 2024

WhisperVoiceTrace (WhiVo) is a novel approach to voice command fingerprinting (VCF) for smart speaker systems such as Amazon Alexa and Google Assistant. By proposing new variants of traffic aggregation matrices (TAMs), WhisperVoiceTrace significantly improves the robustness of deep learning models in identifying voice commands based on network traffic metadata. The study also explores how demographic factors and device location can influence fingerprintability and introduces command category fingerprinting as a scalable approach to identifying unknown commands. This work provides vital insights into the privacy risks posed by voice assistant systems, emphasizing the need for more effective defense mechanisms against VCF attacks.

Recommended citation: Minji Jo, Hyojin Kim, Jiwoo Hong, Hosung Kang, Nate Mathews, Se Eun Oh. (2024). "WhisperVoiceTrace: A Comprehensive Analysis of Voice Command Fingerprinting." ACM Asia Conference on Computer and Communications Security (ASIA CCS), 2024. DOI: 10.1145/3634737.3657017.
Download Paper