Classroom Audio Analysis for Estimating Engagement of the Students

We propose an innovative approach for accurately classifying classroom audio into monotonous and engaging categories using Mel frequency cepstral coefficients (MFCC) feature extraction, and Convolutional Neural Networks (CNN). By leveraging MFCC, our method effectively captures the spectral and temporal information of audio signals. A CNN architecture is employed to classify the audio, utilizing fixed frame lengths and hop sizes for precise interval analysis. The model is trained on a comprehensive dataset and optimized using the Adam optimizer. Extensive experimentation demonstrates the efficacy of our approach for categorizing audio segments and has ramifications for assessing classroom engagement with a score of 58.74%, thereby indicating an optimistic development in teaching practices.