Fuzzy Evaluation System: Intelligent Speech Assessment System

Abstract

This research shows the design of a fuzzy evaluation approach in an intelligent speech assessment system for Malay vowels. The suggested system is really suitable for the children patient can undertake assessment. The target client is 7 to 12 years Malay children. Hidden Markov Model is applied as a technique of speech recognition in the system. Clients are able to exercise and test on their own convenience time independently and the system is able to evaluate their pronunciation as well as the quality of their pronunciation. The fuzzy evaluation session can provide a feedback on the performance of each client vowel pronunciation. In order to have an accurate evaluation, three parameters are involved in fuzzy expert decision making system as inputs. The distance of the formant frequencies, log-probability of HMM and number of correct recognition by HMM play the main rule in vowel quality uttered by client which involved in decision making process of the proposed system. The result of project shows that, the accuracy of the proposed fuzzy expert evaluation system increase up to 89%.