What a SHAME: Smart Assistant Voice Command Fingerprinting Utilizing Deep Learning
Published in 20th Workshop on Privacy in the Electronic Society (WPES), 2021
This work presents SHAME (Smart Home Assistant Malicious Ensemble), a new state-of-the-art voice command fingerprinting (VCF) attack that leverages deep learning to infer user commands from encrypted traffic metadata. The attack is evaluated on multiple datasets, including voice commands issued to Amazon Echo and Google Home, achieving accuracy of up to 99.81% on Google Home and 95.2% on Amazon Echo traffic. The paper demonstrates that voice assistants leak significant information through traffic metadata, posing serious privacy concerns for end-users.
Recommended citation: Jack Hyland, Conrad Schneggenburger, Nick Lim, Jake Ruud, Nate Mathews, Matthew Wright. (2021). "What a SHAME: Smart Assistant Voice Command Fingerprinting Utilizing Deep Learning." 20th Workshop on Privacy in the Electronic Society (WPES), 2021. DOI: 10.1145/3463676.3485615.
Download Paper | Download Slides