Exploiting peer-to-peer communications for query privacy preservation in voice assistant systems

Area of Science:

• Computer Science
• Information Security
• Human-Computer Interaction

Background:

• Voice Assistant Systems (VAS) collect user data, leading to continuous profiling and privacy concerns.
• Current VAS architectures link voice queries to user accounts, compromising source anonymity.
• The increasing prevalence of VAS in daily life necessitates robust privacy-preserving solutions.

Purpose of the Study:

• To propose a novel Voice Assistant System (VAS) anonymizer designed to enhance source anonymity.
• To develop a system that effectively mixes queries from multiple VAS users, obscuring individual query patterns.
• To implement real-time anonymity evaluation mechanisms to mitigate privacy risks.

Main Methods:

• A pattern-matching scheme enables VAS devices to discover peer relays without revealing query patterns.
• Anonymity evaluation modules assess single queries in real-time, reducing pattern violation risks.
• A coordinated query uploading process ensures that service providers cannot link queries to individual users.

Main Results:

• Experiments demonstrate the efficiency of the pattern-matching scheme in terms of computation and communication overhead.
• Anonymity evaluation modules effectively protect the privacy of both query requesters and relays.
• The proposed VAS anonymizer successfully anonymizes query sources, preventing user profiling.

Conclusions:

• The developed VAS anonymizer provides an effective solution for enhancing user privacy in voice assistant systems.
• The system's design balances anonymity requirements with efficient system operation.
• Further research can explore advanced anonymization techniques for interconnected smart devices.