Analysis of Classroom Interaction Using Speaker Diarization and Discourse Features from Audio Recordings.

In this paper we present the rationale, and some initial results, of an automated system for classroom analysis which is based on speaker diarization techniques and non-verbal discourse features extracted from audio recordings. We have employed several Machine Learning algorithms and audio processing methods with classroom recordings related to several undergraduate courses. After determining the identity of the speaker in a recorded class, we can distinguish whether the speaker is a teacher, a student, there are multiple speakers at the same time, or silence. An important contribution of our work is that, from that information, we derive several non-verbal features that can be used to describe patterns. Our preliminary results show that it is possible to extract valuable information using data visualization. As we show, different teachers and teaching methods generate identifiable patterns, that might be used to analyze, for example, which methodologies and teaching styles provide higher levels of interaction or participation.