Few-Shot Audio Classification Model for Detecting Classroom Interactions Using LaSO Features in Prototypical Networks
Year: 2024
Type: conference-paper
Abstract: This research introduces an innovative approach to few-shot sound classification applied to classroom sound recordings that integrates Label Set Operation (LaSO) features with Prototypical Networks. Traditional audio classification methods often require extensive labeled datasets, which can be impractical in real-world scenarios where obtaining large amounts of labeled audio data is challenging. This is particularly the case for the target application of automatically annotating long recordings of classroom audio to understand student learning in classrooms. This paper proposes an enhanced few-shot learning approach based on Prototypical Networks by incorporating LaSO features, to augment the feature space for the Prototypical Network. This methodology focuses on detecting and classifying teacher and student voices for future understanding and analysis of classroom interactions. Experimental results indicate the proposed approach incorporating LaSO features significantly improves classification accuracy of a prototypical network used for few-shot learning. This work paves the way for more advanced and automated solutions in educational environments, facilitating better monitoring and understanding of classroom dynamics.
Authors: Rashed Iqbal, Christian Ritz, Jack Yang, Sarah Howard
Institution: University of Wollongong
Language: English
Topic: Music and Audio Processing
Subfield: Signal Processing
Field: Computer Science
Domain: Physical Sciences
SDG: Quality education
Open Access status: green
OpenAlex ID: W4406858736