Automatic Speaker Verification Spoof Detection and Countermeasures Using Gaussian Mixture Model

Automatic Speaker Verification (ASV) is an emerging biometric authentication technique with the process of accepting/rejecting the users' claimed identity based on his/her speech samples. Robust countermeasures for spoofing attack detections are required to secure biometric systems from intruders. Anti-spoofing is also called replay detection in which voice is recorded, stored and replayed to deceive ASV systems. The ASVspoof series of challenge provides a shared anti-spoofing attack, ASVspoof 2019 focused on both synthetic and replay speech that are referred to as physical and logical access attacks, respectively. To build the robust system, we considered separate data for bonafide and spoofed voice data and implemented separate models for both classes. We addressed our system based on Gaussian Mixture Model, which is performed on ASVspoof 2019 Database. Finally, the experiments focused on both MFCC features and machine learned features have a comparable results with an equal error rate (EER) of 5.64% and 7.56 %.