Research On The Pronunciation Quality Evaluation Model For English Reading Speech

Language is the most convenient means of human communication.With the deepening of globalization,more and more people want to master a foreign language other than their mother tongue.English,as a global language,has attracted more and more attention,among which oral English learning is very important.For English learners,one-to-one,teacherstudent communication and guidance is the most effective way to improve their oral English level,but the traditional English classroom teaching is difficult to meet the oral English learning needs of most learners.Computer Assisted Language Learning(CALL)system,powered by the rapid advancement of speech processing technology,is becoming more and more intelligent,which can not only point out learners’ pronunciation errors,but also evaluate their overall pronunciation level at the same time.This paper studies the two key technologies of pronunciation error detection and pronunciation quality evaluation in CALL system,and combines them effectively to build a pronunciation quality evaluation model suitable for Chinese learners of English.The main research contents of this paper are abstracted as follows:1.Two different methods are studied in pronunciation error detection.The first method is based on the speech recognition framework,the pronunciation accuracy score is calculated for each phoneme and then compared with the phoneme independent threshold to detect pronunciation errors.In this method,the standard American English acoustic model is adjusted to match the English pronunciation of Chinese students,and the phoneme dependent threshold is set to improve the performance of pronunciation error detection.2.In view of the fact that the first pronunciation error detection method has limited ability to model easily confused phonemes,the second pronunciation error detection method based on the knowledge of acoustics and phonetics is studied.This paper mainly focuses on the typical easily confused phoneme pair /i/and /(?)/ that are frequently mispronounced by Chinese students.Many discriminative acoustic-phonetic features are extracted from these two phonemes and then used to train and test the Support Vector Machine(SVM)classifier,the experimental results show that the proposed method can achieve better performance than that of the first method.3.In order to evaluate students’ overall pronunciation quality more comprehensively,three different dimensions are evaluated respectively,which includes pronunciation accuracy,pronunciation fluency and intonation.In the evaluation of pronunciation fluency,a new evaluation feature is designed,that is,the word duration ratio.In the evaluation of intonation,the method of intonation evaluation based on Dynamic Time Warping(DTW)algorithm is studied.Finally,the Support Vector Regression(SVR)algorithm is introduced to combine all these evaluation features of different dimensions into an ultimate pronunciation quality score,which significantly improves the performance of pronunciation quality evaluation.