Design And Implementation Of Pronunciation Automatic Assessment System

The automatic assessment of pronunciation is an integral part of Computer-Assisted Language Learning (CALL), the aims of which are to carry out automatic assessment of the language learners' pronunciation capability, help them adjust their pronunciation and, ultimately, improve their oral English capabilities. However, the influence of their mother tongue on their pronunciation capability for a secondary language and the intrinsic characteristics of an individual's pronunciation, coupled with other common elements, such as the speech speed and tone, have hindered the accuracy and effectiveness of the assessment. The common method adopted to solve this problem is to establish a language model using automatic speech recognition technology to recognize the pronunciation under test and evaluate the pronunciation capabilities. To some extent, the automatic assessment method for pronunciation has been brought to maturity by the development in automatic speech recognition.After systematically exploring the fundamentals of Hidden Markov Model(HMM), an English pronunciation model based on HMM is introduced and specifically adjusted in accordance with the characteristics of Chinese-speaking learners. In the establishment of this model, the selection and extraction of speech feature parameters and the training method for parameters of the model are also explained in detail. Based on the proposed model, an algorithm for the classification of pronunciation capabilities is also presented and an implementation of automatic assessment system for pronunciation is also introduced at length.In the proposed system, the influence of mother tongue pronunciation on the pronunciation of secondary language has reduced the similarity between assessment model state and pronunciation under test and decreased the accuracy of the proposed system. To fix this problem, an updated version of the proposed model is obtained through adjusting the original model by taking into account the syllables that can be easily misunderstood for Chinese-speaking English learners. To assess the performance of the proposed system, an evaluation algorithm for the proposed system is presented, in which the distance between Viterbi forced alignment coefficient and automatic speech recognition coefficient is used as the yardstick. The pronunciation capabilities of the testees then can be correctly assessed by mapping the received coefficients into score range.In this document, an automatic assessment system for pronunciation based on HTK is presented. The system is exclusively designed for English learners whose mother tongue is Chinese. The proposed system is mainly composed of three modules, namely, pronunciation assessment engine, voice-capturing and processing module and the display module. The core module of the system, which is the pronunciation assessment engine, is developed using HTK. The data transmission between different modules can be achieved through label files. Experimental results have shown that the correlation between scores obtained by the proposed system and scores obtained manually is 0.89, which illustrates the effectiveness and accuracy of this system.