Towards protecting cyber-physical and IoT systems from single- and multi-order voice spoofing attacks

摘要

Voice-controlled systems (VCSs), a new class of cyber-physical systems (CPS), and Internet of Things (IoT) devices are increasingly employing smart speakers such as Google Home and Amazon Alexa, and other voice assistants to enable management of various remote operations at home and offices. However, these smart speakers and hence VCSs are susceptible to various voice spoofing attacks i.e. replay, cloning, etc., in a non-network environment as well as in a multi-hop network setup. These diverse spoofing threats on VCSs require an urgent need to develop a robust spoofing countermeasure for VCSs capable of detecting a variety of voice spoofing attacks. This paper presents a spoofing countermeasure that uses novel acoustic ternary patterns (ATP) with Gammatone cepstral coefficients (GTCC) features to counter the voice spoofing attacks on VCSs in single- and multi-hop network environments. Our experimental analysis demonstrates that the proposed ATP features when combined with GTCC can effectively detect the distortions in replayed samples, unnatural prosody present in the cloned samples, and both distortions and unnatural patterns of stress and intonation in cloned-replay samples. The proposed ATP-GTCC features are used to train the SVM for development of a spoofing countermeasure to cater all possible forgeries. Experimental results based on highly diversified ASVspoof 2019 and VSDC datasets signify the effectiveness of the proposed countermeasure for reliable detection of 1st- and 2nd-order replay, cloning, and cloned-replay attacks. (C) 2021 Elsevier Ltd. All rights reserved.