Voice Based Authentication System for Web Applications using Machine Learning

Due to security concerns, the biometric trend is being used in many systems. Biometric authentication is a cheap, easy, and reliable technology for multi-factor authentication. Cryptosystems are one such example of using biometric data. However, this could be risky as biometric information is saved for authentication purposes. Therefore, voice biometric systems provide more efficient security and unique identity than commonly used biometric systems. Although, speech recognition-based authentication systems suffer from replay attacks. In this paper, we implement and analyze a text-independent voice-based biometric authentication system based on the randomly generated input text. Since the prompted text phrase is not known to the speaker in advance, it is difficult to launch replay attacks. The system uses Mel-Frequency Cepstrum Coefficients (MFCC) to extract speech features and Gaussian Mixture Models (GMM) for speaker modeling.