Improvement of Speaker Verification Using Deep Learning Techniques

Abstract

AbstractDeep learning (DL) has been used to solve a range of real-time artificial intelligence (AI) challenges with great success. This is a cutting-edge area of machine learning (ML) that has been rapidly evolving. As a result, deep learning is quickly becoming one of the most popular and well-defined machine learning techniques, with applications in a wide range of fields, including image processing, computer vision, speech and speaker recognition, emotion recognition, natural language processing, hand-written character recognition, cyber-security, and many others. Over other prevalent methods, DL approaches have demonstrated superior performance in speech processing areas like as voice recognition and speaker recognition. We describe an experimental setup for speaker verification (SV) utilizing DL techniques, and discuss its performance and findings, as well as how it outperformed established approaches such as HMM, GMM-UBM, and SVM. In this research works, we analyse and review deep neural network (DNN) approaches employed in SV systems. With a 1.51% equal error rate (EER), the final result is the best performance of the SV systems of restricted Boltzmann machine (RBM)-based DNN.KeywordsSpeaker verificationMachine learningDeep learningDeep neural networkMFCC