Accurate, Fast and Low Computation Cost of Voice Biometrics Performance using Model of CNN Depthwise Separable Convolution and Method of Hybrid DWT-MFCC for Security System

Abstract

Identity theft presents a substantial criminal threat in the digital world, especially in online transactions. To overcome this problem, voice biometrics was created as a method to guarantee identity security. This research is to look at voice biometrics systems that use deep learning model, focusing on the CNN Depthwise Separable Convolution (DSC) model compared to CNN Residual. The comparison of these two systems is to improve accuracy and performance. CNN Residual's first Voice Biometrics testing showed a high accuracy validation performance 98.6345%. The large number of Residual CNN training parameters causes a longer training process time 7.37 seconds and response time 2.35 seconds. So, the computing load becomes larger. The second voice biometrics test of CNN DSC showed high accuracy validation performance results 98.3542%. CNN DSC performance succeeded in reducing the number of training parameters, thereby shortening the training process by 5.12 seconds and the fastest response time was 1.54 seconds. Based on the analysis of the test results above, it shows performance advantages. CNN DSC is able to reduce the computing load, is able to improve the user identity security system in banking transactions accurately and quickly and is able to solve the problem of high computing costs.