Publication Details
Resilience of Voice Assistants to Synthetic Speech
Firc Anton, Ing. (DITS FIT BUT)
Kaška Petr, Bc. (FIT BUT)
Lapšanský Tomáš, Ing. (DITS FIT BUT)
Šandor Oskar, Bc. (FIT BUT)
Homoliak Ivan, doc. Ing., Ph.D. (DITS FIT BUT)
Deepfake; Security Analysis; Spoofing Attacks; Voice Assistant; Voice Biometrics
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.
@INPROCEEDINGS{FITPUB13123, author = "Kamil Malinka and Anton Firc and Petr Ka\v{s}ka and Tom\'{a}\v{s} Lap\v{s}ansk\'{y} and Oskar \v{S}andor and Ivan Homoliak", title = "Resilience of Voice Assistants to Synthetic Speech", pages = "66--84", booktitle = "Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)", year = 2024, location = "Bydgoszcz, PL", publisher = "Springer Nature Switzerland AG", ISBN = "978-3-031-70878-7", doi = "10.1007/978-3-031-70879-4\_4", language = "english", url = "https://www.fit.vut.cz/research/publication/13123" }