Research On Voice Transformation Spoofing Detection Algorithm And Implementation Of Robust ASR System

It has been revealed by the existing researches and practices that speech spoofing including voice transformation(VT),voice conversion(VC),speech synthesis(SS)and replay attack can deceptive the state-of-art automatic speaker recognition(ASR)systems and presents serious threats to social security.Reported efforts mainly focus on the detection of VC,SS and replay attack while the researches on VT-related security issues are relatively insufficient.However,the VC and SS algorithms are not well developed and the cost is high,while the existing VT algorithms can change one's speech with acoustic naturalness in a simple and low-cost way.VT has been integrated in many prevailing audio processing tools and employed in various criminal cases.Therefore the research on VT spoofing security issues is of great significance.In this thesis,we study the detection algorithm of VT spoofing detection,and develop a robust ASR system against VT spoofing.The main contributions of this thesis is as follows.To propose a VT spoofing detection algorithm based on Gaussian Mixture Model-Universal Background Model super vector derived from MFCC.SVM(Support Vector Machine)is used as the classifier.The experimental results show that the accuracy rates achieve over 98.62%.In addition it can maintain a high accuracy with short testing segments.To develop a robust ASR system against VT spoofing in Visual Studio 2013 MFC framework.The spoofing detection module of this system is based on the proposed algorithm mentioned above and the other one proposed by our lab team,and it consists of 3 sub-modules,i.e.,the feature extraction module,the model training module,detection and the decision-making module.This module is integrated into a GMM-based ASR system.Testing has conducted to reveal that the system can detect VT spoofing effectively to maintain a standard and normal false reject rate.The system has a friendly human-machine interaction interface,and it is practical.It has a functional operation interface and a console window during operation,which is convenient for users to get real-time feedback when the system is running.To summarize,in this thesis we research on the VT spoofing detection algorithmand develop a robust ASR system,which has great significance to the research of spoofing security issues.