| In recent years,with the rapid development of technology,automated oral test can be used to evaluate the type of reading questions more reliably,it can replace part of manual scoring and assist non-professional teachers to make more reasonable scores However,the affective index has been ignored in oral test.This study focuses on the construction of the automatic scoring model of the pronunciation quality of Chinese reading questions.and collect the reading speech data of the fourth grade students,looking for a method that can measure the learners' emotional pronunciation quality,and integrates it into the comprehensive evaluation model of the pronunciation quality.According to the "three character analysis method" established by the academic community to measure the oral performance of learners,and combined with the characteristics of this study,based on the third criterion of reading aloud passage in Putonghua proficiency test,the scoring criteria of this study are formulated to determine the scoring characteristics of reading questions.The scoring characteristics of the automatic scoring of reading questions are divided into three dimensions: accuracy evaluation,fluency evaluation and emotion evaluation.In the aspect of accuracy evaluation,the ful-scale score of reading question type is obtained by calling i FLYTEK's voice evaluation interface,and the average score of rhyme score,tone score and integrity score is selected as the intonation score.Because of the reason of i FLYTEK's scoring mechanism,the fluency score does not meet the scoring standard proposed in this study,so the fluency given by the evaluation interface is discarded In turn,the endpoint detection technology based on short-time energy and zero crossing rate is used to detect the mute part of the answer voice,cut out the total effective pronunciation time,and compare with the total pronunciation time of the standard voice,and the score obtained is taken as the speech speed score.In the same way,count and count the number of silence segments to get the number of pauses.The ratio of the number of pauses with the standard pronunciation is used as the pause score.In the aspect of emotional evaluation,based on the current research results of emotional features,the data of speech signals such as energy,fundamental frequency,resonance peak,etc.are extracted as the emotional features of speech.The similarity of standard speech and test speech on these feature vectors is calculated by cosine similarity,which is the emotional expression score of ancient poetry reading.Finally,the average score given by the three raters is taken as the dependent variable,and the machine score of each dimension is taken as the independent variable.The multiple stepwise linear regression equation is constructed by SPSS software,and the final automatic scoring model of reading questions is as follows:Score = 1.278 + 0.569 * emotion score + 1.496 * rhythm score + 0.558 * speech speed score + 0.249 * intonation scoreThrough the correlation test between the predicted score from the scoring model and the original score from the manual,it can be found that the consistency rate,adjacent consistency rate and Pearson correlation coefficient of human-computer scoring are relatively high,which proves that the scoring model constructed in this study has a good effect.At the same time,the score model with emotional indicators is higher than that without emotional indicators in the correlation of human-computer scores,which also verifies the effectiveness of the emotional indicators proposed in this study. |
PDF Full Text Request