An Acoustic Analysis System For Classroom Teaching Assessment

The traditional way to evaluate the quality of teaching is inviting experts to sit in the class and rate according to several criteria.The purpose of this thesis is to establish a speech analysis system,to evaluate teaching automatically,and to help teachers improve their teaching.The speech analysis system includes voice acquisition,preprocessing,feature extraction,VAD,speaker recognition and scene classification.We define five common teaching scenes:teachers' speech,students' answer,reading aloud,discussion and mute.We use GMM,GMM-UBM and i-vector to classify scenes,and compare the results of different models on two video sources,real classroom recording videos and high quality videos from Internet.We then identify appropriate models,build database of teaching scenes and mark some courses for further research.Based on the scene classification,we try to assess teaching from three aspects: the speech speed,the voice loudness and frequent scene sequences.The speech speed and voice loudness are estimated from the speech fragments of teachers' speech.Then all the classified scenes can be connected into a series of scenes in sequence.We conducted an exploratory data analysis with these scene sequences.The procedures are as follows: summary statistics,frequent scene sub-sequence discovery and process mining.The process mining method is used to discover the interactive mode of the courses.By analyzing the process results,we can study the characteristics of the courses and teachers,which can be useful information for teachers to improve their teaching.