Investigation of the performance of high-frequency phase spectrum components in audio replay spoofing attack detection
Burak Kasapoglu, Turgay Koç · 2019
Automatic speaker verification (ASV) systems are being used in various areas such as mobile banking, security, and military. Due to widespread use of ASV systems, spoofing attacks are developing themselves and getting more important problem for ASV systems. In this study, we investigated currently proposed features to solve this problem and proposed the usage of high-frequency components of phase spectrum based feature modified group delay (MGD) for audio replay spoofing attack detection. Then, spoofing detection performances of magnitude spectrum based feature constant Q cepstral coefficients (CQCC) and high frequency components of CQCC are compared with high frequency MGD performance. We used the ASVspoof 2017 database to test our countermeasure system. As a result, high-frequency components are performing %40-55 better performance than low frequency for both magnitude and phase spectrum based features and it improves the baseline system performance which is proposed by ASVspoof 2017 by %40-%45 in total over the development sub-dataset.