Replay Attacks Detection Using Phase and Magnitude Features with Various Frequency Resolutions

Replay attacks detection is a challenging task under unseen recording environments and playback devices. To overcome this problem, we propose a combination of phase-based and magnitude-based feature with various frequency resolutions. Conventional feature used in replay attacks detection does not pay much attention to phase which is expected to be effective for replay attacks detection because the genuine and spoof speech has significant differences in phase. In this paper, we investigate the combination of phase-based and magnitude-based features with different frequency resolutions to better solve the classification issue. The replay attacks detection results on ASVspoof 2017 challenge indicated that our proposed approach achieved 51.3% relative error reduction rate than the conventional magnitude-based feature.