The optimal algorithm of sub-symptom threshold exercise training for aural habilitation/rehabilitation

Abstract

This paper presents the comparison study of the speech recognition system for the Thai language in the noise of the different environment. The well-known algorithms, such as MLP, SVM, GMM, HMM, VQ, DTW, DNN and End to End were used in this research. A test was conduced with 50 men and 50 women subjects during 5-60 years old. The proposed method consists of several parts which are (i) the feature extraction by Mel-frequency cepstral coefficients (MFCC) algorithm, (ii) The learning and decision process. The performance testing of the systems by the Ling's six sounds, such as ah, mm, oo, ee, sh and ss. The experiment results of our proposed method show that the accuracy of the system more than 80 percent. © 2017 Association for Computing Machinery.