The Development of Thai Language Pronunciation Training Software Focused on Words Spelling with ร-ล using a Hidden-Markov Model

Abstract

This paper presents the development of pronunciation training software aimed for solving the pronunciation problems of Thai words with ร-ล initial consonant i.e. pronouncing the ร without trill and pronouncing the ล with trill. Training is divided into seven chapters, each consisting of a lesson, exercises and a game. A lesson demonstrates the correct lip-movement of training words by means of animations and video. The exercises and game evaluate the student’s speech obtained via a microphone, and the pronunciation ability results are reported immediately. An 11-state 8-mixture Hidden-Markov model recognition engine was utilized to evaluate the students’ speech. When the software was tested with 6 second grade primary school students, the average pre-test and post-test scores increased from 41.27 to 78.57, and the E1/E2 efficiency was found to be 82.14/78.57