Enhancing Security of Automatic Speaker Verification System

Abstract

Speaker Verification is like biometric system in which it is verifying speakers identity. Here, Robust Speaker Verification is designed. The use of this robust system is mainly for security issues. The main aim of this system is to authenticate the correct speaker and make secure the system from fake or spoofed speakers. Here, speech signal is used as input of the system. This system requires pre-processing like Feature Extraction techniques such as MFCC, LPC, or any other features. Classification is used for build a model and then feature matching is requires. Decision making is done after features are matched. Researches show that biometric technology can be innocent to harmful spoofing attacks. There are many spoofing attacks are very famous such as impersonation, replay, speech synthesis and voice conversion. Speech synthesis is recent issue in this Field so this verification system is made against speech synthesis attack. Here, by means of detecting spoofing attacks we can improve the performance of the automatic speaker verification system very easily. As per our proposed work, here we used two different feature extraction techniques which are MFCC and LPC. Both features are matrix. After feature extraction Support Vector Machine classification algorithm is used. Experimental results shows that fusion of MFCC and LPC get 58% accuracy in K=10 fold validation for speech synthesis dataset. Only MFCC is also getting 61% accuracy but there is no such difference between both method's accuracy. If we extracted more features then we will get better results for Speaker Verification System.