Research on Automatic Speech Analysis for Teacher's Q&A in Classroom

摘要

Question-and-aswer (Q&A) is the important content of classroom evaluation. In order to meet the needs of large-scale class evaluation, this work uses voice analysis technology to implement exploratory research on automatic Q&A analyse, mainly on speaker recognition based on MFCC Gaussian mixture model and closed/open question identification based on the logistic regression method. The Q&A auto-analysis system developed in this work can perform automatic analyses for several classroom evaluation indexes including Q&A times, speaking durations of the teacher and his students, and the number of open questions and closed questions. For the real classroom teaching video down-loaded from "CCTV network Chinese public class", both the speaker recognition and closed/open questions identification of this work have obtained satisfied recognition accuracy with recognition rates above 93%.