Automatic speaker verification (ASV) is the process of verifying a claimed speaker identity from a voice signal based on speaker-specific characteristics. However, it has been widely acknowledged that a generic ASV system may be attacked. Urgent demands for anti-spoofing countermeasures are suggested by numerous vulnerability studies. The main focus of this thesis is to propose high-performing anti-spoofing solutions designed from various perspectives in spoofing detection. This work contributes to generate insights of applying state-of-the-art approaches for developing efficient anti-spoofing systems. The thesis deals with the problems of feature optimization and system generalization, insufficient and imbalanced training data and integrated spoofing robust ASV systems.
|Qualification||Doctor of Philosophy|
|Award date||9 Apr 2020|
|Publication status||Unpublished - 2019|