Research On English Pronunciation Quality Autoscoring Fusing Speech Emotion And Design Of Application System

Speech emotion is a kind of important information in speech signal.In language ?Input – Output‘ acquisition model,Speech emotion is also an important input and output elements.Although there have been many achievements in the field of English pronunciation quality auto-mated evaluating technology,but most studies focus on pronunciation,fluency,rhythm,intonation and other indicators.Few scholars take speech emotion into consideration as an indicator of which are individually assessed.Faced to the situational oral teaching and pronunciation quality assessment,the existing evaluation methods are clearly limited.Speech Emotion automatic identification technology,is the voice signal processing and machine learning combines popular subjects in sentiment classification,corpus building process,feature extraction,recognition method and application,which have to learn the outcome.In this context,this paper will combining the pronounce quality evaluation and speech emotion,studies the speech emotion feature selection,speech emotion pronunciation quality evaluation methods,multi-index comprehensive evaluation method of pronunciation quality,application system design and other issues.The main research work includes:(1)Analysis of the contribution of the different characteristics of speech emotion.This paper preliminary selected 36 covering energy,fundamental frequency,formant,slope of pitch tail and speech rate,as the speech emotion feature.Using principal component analysis,optimized for feature extraction cumulative contribution rate of 95% of the first 19 principal components,and calculate the original features of the main components of linear contribution.(2)Design of the speech emotion pronunciation quality evaluation method.In this paper,?recognition? was transfer to ?evaluation? by the use of soft classification characteristic classifier.Support Vector Machine(SVM)is used as effective speech emotion recognition.This paper takes the output value of the confidence probability of SVM as a result of the evaluation of the measure.Faced to the difficult problem of selecting the SVM kernel function and parameter,this paper applied swarm algorithm to the parameter optimization.Taking into account the evaluation data corpus imbalances,to inadequate sampling method,based on multiple classifiers bagging the mean as the evaluation result.By the mean and variance of the data analysis,and the use of one-way ANOVA are verified herein emotional evaluation method is effective.(3)Design of multi-index pronunciation quality evaluation method based on Decision Tree.The traditional method of multiple linear regression,and does not apply to corpus application scenarios of this research.This paper presents a decision tree structure,similar to the overall rating process raters.Use ID3(Interactive Dicremiser version3)decision tree algorithm to construct a comprehensive evaluation of pitch,rhythm,intonation,speed and emotion five indicators.The result of experiment showed that agreement is 73.9%,adjusted agreement is 93.8%,Pearson correlation coefficient is 0.81.(4)Develop a spoken language learning system.Learners can watch and imitate the reading materials.The program evaluates the learners‘ reading speech.Teachers can diversify the statistical results,to understand students' oral level.