Automated classification of classroom climate by audio analysis

While in training, teachers are often given feedback about their teaching style by experts who observe the classroom. Trained observer coding of classroom such as the Classroom Assessment Scoring System (CLASS) provides valuable feedback to teachers, but the turnover time for observing and coding makes it hard to generate instant feedback. We aim to design technological platforms that analyze real-life data in learning environments and generate automatic objective assessments in real-time. To this end, we adopted state-of-the-art speech processing technologies and conducted trials in real-life teaching environments. Although much attention has been devoted to speech processing for numerous applications, few researchers have attempted to apply speech processing for analyzing activities in classrooms. To address this shortcoming, we developed speech processing algorithms that detect speakers and social behavior from audio recordings in classrooms. Specifically, we aim to infer the climate in the classroom from non-verbal speech cues. We extract non-verbal speech cues and low-level audio features from speech segments and train classifiers based on those cues. We were able to distinguish between positive and negative CLASS climate scores with 70–80% accuracy (estimated by leave-one-out crossvalidation). The results indicate the potential of predicting classroom climate automatically from audio recordings.