Research On Teacher's Speech Emotion Recognition Based On Deep Learning

Emotional expression is an important component of human communication and interaction.The ability to recognize,understand,and express emotions given to machines through speech emotion recognition technology is the future development direction of artificial intelligence.In recent years,the rapid development of speech emotion recognition in emotion computing,learning emotion detection,mental health analysis,customer service detection,etc.has attracted much attention.In the field of artificial intelligence education,teacher teaching discourse is the main method of classroom teaching,and teaching emotion,as the main evaluation basis for teaching evaluation,has a profound impact on teachers' teaching methods,classroom atmosphere and teaching effects.In the face of massive teaching data,traditional manual methods such as classroom observation,questionnaires and teacher interviews are mainly used for teacher emotion judgment and classroom atmosphere analysis.Therefore,how to use deep learning technology to automatically recognize teachers' speech emotions and realize intelligent analysis of classroom atmosphere is the main problem faced by teaching evaluation under education big data.Aiming at the current lack of intelligent evaluation methods for teacher emotion in the field of artificial intelligence education,this paper proposes a teaching speech emotion recognition model that combines frame-level features and discourse-level features.This model combines frame-level features and discourse-level features,and conduct speech emotion recognition experiments on open source databases and teaching speech databases,and finally apply them to real classrooms,using the teaching speech emotion recognition system to visualize teaching emotions and classroom atmosphere.The experimental results show that this method has a good recognition effect compared with other mainstream methods in the task of speech emotion recognition.The main work and innovations of this paper include the following three aspects:First of all,this paper designs different speech emotion feature extraction modules based on the temporal and spatial characteristics of speech.For the frame-level temporal features and speech-level global features of speech,two types of speech based on frame-level feature representation learning and utterance-level feature representation learning are designed respectively.Emotion recognition model,and conducted a speech emotion recognition experiment on the open source speech database.The experimental results show that the two feature representation models have their own advantages in emotion recognition tasks.Secondly,this paper combines the advantages of the spatial and temporal features of speech,and designs a LSTM-FCN-LSTM parallel network(LFL-Net),in which the LSTM-FCN network extracts the spatial and temporal characteristics of the speech signal,and the LSTM network extracts The fusion of the global features of the speech signal improves the characterization ability of the features.The experimental results show that the network model based on LFL-Net is tested in the open source speech emotion library,and a good recognition effect is achieved.Finally,based on the traditional classroom teaching scene,we collect the speech data of the teaching scene in the middle school classroom,preprocess the speech data including noise removal,noise reduction,and mute cutting,and then perform manual proofreading and labeling,and establish the teaching voice emotion corpus TSEC.The LFL-Net model trained on the corpus is used to realize the automatic recognition of teachers' speech emotions in teaching,and visually display the emotions of classroom teachers and the emotional atmosphere of classroom teaching.