文章基本信息

标题：DYNAMIC TIME WARPING AND HIDDEN MARKOV MODEL CLASSIFIER IN TEXT DEPENDENT SPEAKER VERIFICATION SYSTEM WITH DIFFERENT SPEAKING VARIANT
本地全文：下载
作者：KEVIN KURNIAWAN ; RAYMOND SAMUEL ; BENFANO SOEWITO 等
期刊名称：Journal of Theoretical and Applied Information Technology
印刷版ISSN：1992-8645
电子版ISSN：1817-3195
出版年度：2021
卷号：99
期号：19
语种：English
出版社：Journal of Theoretical and Applied
摘要：One of the main focus on IoT is how to make the facility more secure in order to increase the safety and convenience of the facility. While traditional authentication approach for verifying identity such as Personal Identification Number (PIN) or password is widely used, traditional authentication approach have major flaws that can make the verified user lost access to the protected resource or even pose a major security breach. Speaker verification is one of the many biometric system solutions to solve that problem. In speaker verification, a different speaking pattern and loudness of the voice may affect the performance for the system to verify an individual identity, which a change of speaker speaking pattern will occur very likely since a speaking pattern is highly affected by the speaker�s mental and physical condition, and loudness of the voice is affected by how far the speaker away from the microphone. In this paper, we compare two well research text-dependent speaker verification methods, dynamic time warping and hidden markov model, on verifying user�s identity on different voice variant (loud, normal, quiet, shout, and soft) to replicate the condition when the speaker is experiencing mental or physical conditions and the distance between the speaker and the microphone that affects the loudness of the voice and speaker speaking pattern. This paper uses 330 total train data and 1,600 total test voice data where every test voice data will be retested for every registered user. Research done in this paper shows hidden markov model achieved better accuracy on normal, shout, and soft voice variant by 2,6%, 0,5%, and 0.74% respectively, whereas dynamic time warping achieves better accuracy on loud and quiet voice variant by 2,79% and 2,3% respectively.
关键词：Speaker Verification;Text Dependent Speaker Verification;Hidden Markov Model;Dynamic Time Warpi