Research On Teacher’s Discourse Emotion Recognition Based On Traditional Classroom

As the foundation of teacher professional development,teacher emotion profoundly affects teachers’ self-cognition,classroom teaching effect and student learning effect Speech emotion recognition is an important way to achieve harmonious natural human-computer interaction.This paper applies speech emotion recognition to traditional classrooms,analyzes teachers’ diseourse emotions,helps teachers adjust classroom status in time,and promotes teacher professional development.In this paper,the discrete emotion corpus contains a single kind of emotions,which can not meet people’s needs for complex emotions,and does not match the traditional classroom environment.It mainly performs feature extraction,feature selection,feature fusion,emotion recognition,etc.on the dimensional emotion corpus,and finally get the teacher’s emotional pleasure curve,to provide teachers with a reference.The main work and innovations of this paper are as follows:Firstly,the different feature sets of opensmile are used to extract the CASIA discrete corpus and TAL dimension corpus emotional features.The traditional machine learning method is used to identify and compare with the five types of features extracted by traditional methods.The experimental results illustrate the effectiveness of opensmile in extracting emotional features.Secondly,for the TAL dimension corpus,the global feature and time series features are extracted by using different feature sets of opensmile,and a method for selecting two types of features based on fully connected network(FC) and one-dimensional convolutional neural network(1D-CNN) is proposed.The experimental results show that the combined global feature of feature set IS10+IS13 is better,and the combined timing feature of feature set IS1011d+IS1311d is better.Thirdly,based on the global feature and time series characteristics,a method based on correlation neural network(CorrNet) fbsion feature is proposed to reduce the correlation between global features and time series features.Experimental results show that CorrNet-based fusion features can be used.Get better predictions.Finally,based on the above research,a set of specific models for analyzing teacher,s discourse emotions in traditional classroom environment is designed.The small sample test shows that the model can effectively analyze the classroom teachers’ discourse emotions.