Fraud Recognition Using Voice Authentication Through Deep Learning Applying 1D-CNN Algorithm

Abstract

The identification of individuals has become one of the most significant issues in the protection of any system, this is because of the spread of the Internet, computers, and electronic accounts. The recognition of users through biometric features is one of the fundamental principles, and voice is one of the most effective ways to recognize... Compared to other traditional biometric methods, they offer a similar level of security and are dependable and trustworthy. The objective of this research is to explore deep learning using the proposed one-dimensional convolutional neural network (1D-CNN) on two different voice datasets: natural voice recordings and unconstrained voice recordings. The dataset that has the greatest success in recognizing and categorizing speakers is the voice recording that is taken in a natural environment. To enhance the quality of the acoustics' in real-world environments, we pre-process the audio using noise reduction and enhancements based on Mel-Frequency Cepstral Coefficients (MFCC) for each sound, as well as their variance and acceleration. Our first attempt at training and testing the 1D-CNN, the results showed that it had a 100% success rate.