A Study Of Rating Method Of Intonation In Acoustic Feature-Based English Diagnostic Test

In traditional spoken English proficiency test, the participant's pronunciation quality was judged by raters according to their general impression, which resulted in the subjectivity of rating. What was more, the study was limited to primary intonation system which simply described the intonation grouping or intonation pattern of the participants' speech without concrete calculation of the differences between the intonation of participants'and referenced standard pronunciation. In order to reduce the subjectivity and improve reliability of rating, integrating the application of Praat, this study deeply researches the second intonation system by referencing the analysis of Chinese reading tone and the methods of automatic evaluation of Chinese prosody. The difference of the intonation between the participants'and referenced standard pronunciation is calculated in this thesis and finally three-level ranges are decided. there is the realization of Praat automatic rating of the intonation on the basis of the experts' judging results. An objective judging criterion is formed, which can benefit phonetic test.This thesis starts with analyzing the general pitch tendency, pitch slope variation ratio, and MFCC on the basis of the definition that intonation refers to the pitch pattern in spoken English. The main researching content includes the intonation rating of five interrogative sentences in English. Through both qualitative and quantitative study, a C,A,P judging scale hypothesis is formed and the numerical range of each level is figured out according to both the experts and Praat.The main steps in the experiment are as follows:(1) A corpus of English interrogative sentences is built which are selected from the New Concept English edited by Alexander and He Qixin and published by Foreign Language Teaching and Research Press.(2) Select testing participants who are females because the referenced standard pronunciation is the recording of female and there is acoustic difference between different genders. The speech contest winner,students passing the spoken English test in CET band 4 or 6 and someone with heavy mother tongue accent are involved.(3) A three level judging scale hypothesis is put forward on the basis of former rating scales. Then four expert raters are invited to judge the participants'intonation according to this scale.(4) Divide C,A,P ranges. There are three main factors involved in:general pitch tendency, pitch slope variation ratio, and MFCC. Then the proportion of each factor is determined. Finally, the large calculating results are transferred into 100 mark system.Finally, referencing both the marks and the expert raters'suggestion, the ranges of three levels (correct, acceptable, problematic) are formed.The study shows that this rating method can provide objective reference for pronunciation test,so it overcomes the limitation that different raters have different judging levels to the performance of the same participant. Therefore, it can improve the reliability and efficiency of pronunciation diagnostic test.