Replay Attack Detection in Multimedia of Things (MoT)

Fake audio detection is expected to become an important research area in the field of multimedia of things (MoT). This paper aims to propose a countermeasure to detect replay spoofing attacks. More specifically, using controlled data sets prepared by SMILES (Semantic Modeling and Intelligent Learning in Engineering Systems) Lab and ASVspoof 2019, we compared the performance of several machine learning algorithms and Bidirectional Long-Short Term layer deep learning algorithms. Detailed evaluation shows that the SVM classifier performed the best among other algorithms by achieving a precision of 97.9%, recall of 97.7%, f-measure of 97.8%, and accuracy of 98%, while maintaining high efficiency. Additionally, detailed analyses of feature importance show that MFCC, GTCC, and Spectral Flatness are the most reliable features for differentiating between original, first order replay, and second order replay audio samples.