Contents

-

Intelligent English Classroom Interaction Design and Teaching Method Optimization Based on Deep Reinforcement Learning

Yajuan Zuo1
1Basic Teaching Department, Shanxi College of Applied Science and Technology, Taiyuan, Shanxi, 030062, China

Abstract

At present, the evaluation of spoken English in domestic universities is affected by the evaluation teachers’ personal cognition, preference, time, energy and other factors, and it is difficult to unify the standard of oral evaluation in the implementation, and the evaluation frequency and timeliness are insufficient to meet the students’ willingness to improve their oral language. In this paper, multimodal speech recognition technology is utilized to firstly collect students’ speech signals through microphone arrays, secondly extract acoustic and linguistic features of speech, and construct multimodal feature vectors by combining visual information such as students’ lip movements and facial expressions. Subsequently, the feature vectors are input into a deep neural network model for training and recognition, fusing LSTM network with attention mechanism to analyze the speech emotion and capture the emotional changes in speech. Meanwhile, the interaction behavior in speech is analyzed by combining temporal convolutional network. Construct a deep reinforcement learning model, introduce a user item interaction layer, design a user interaction simulator, and obtain user feedback on the smart English classroom. Using multimodal speech recognition technology, the temporal waveform of classroom speech is analyzed for sound pressure value, and the normalized sound pressure value range fluctuates around [-1.5,1.5].The average recognition rate of the six emotions rises to 67.86% with the joint effect of LSTM and attention mechanism. By comparing the experiment, analyzing the difference between the experimental class and the control class before and after the reading aloud ability, the average score of the experimental class is 23.945, and the average score of the control class is 21.464, at the same time, the post-test of reading aloud ability corresponding to the experimental class and the control class P=0.005<0.05. It can be seen that the intelligent interactive classroom of English language constructed in this paper has a facilitating effect in the process of teaching reading aloud in the aspect of reading aloud ability of students The classroom can be seen that the intelligent English interactive classroom constructed in this paper has a promoting effect in the process of teaching reading aloud in terms of students' reading ability.

Keywords: multimodal speech recognition, deep neural network model, LSTM network, user interaction simulator, English interactive classroom