Resilience of Voice Assistants to Synthetic Speech

With the increasing integration of voice assistants in smart home systems, concerns regarding their security, especially regarding personal information access and physical entry control, have escalated. This is further amplified by the rapid development of generative AI methods, which bring new types of attacks. Therefore, we focus on modern voice assistants and their resilience against deepfake spoofing attacks. We rigorously assess the resistance of smart devices to sophisticated audio impersonation techniques. In detail, we evaluate voice assistants on four devices (Google Assistant, Siri, Bixby, and Alexa) with 72 test subjects. Subsequently, we conduct a comprehensive security analysis to determine the extent of potential impacts stemming from identified vulnerabilities. Our findings contribute to the enhancement of voice assistant security, ensuring safer and more reliable utilization in domestic environments.