Semi-supervised Learning for Acoustic Scene Classification using FixMatch

Acoustic scene classification (ASC) plays a critical role in enhancing sound event detection by providing necessary context. Given the diversity of acoustic environments in different cities and countries, a significant amount of data is required to build robust ASC systems. Additionally, ASC must perform reliably on various recording hardware, including microphones on mobile devices. However, the annotation of such extensive datasets is both costly and impractical. Semi-supervised learning (SSL) offers a viable solution by incorporating unannotated data into the learning process. In this paper, we present the current state-of-the-art in semi-supervised learning for ASC, with a particular focus on the FixMatch algorithm. We evaluate FixMatch on several public datasets using varying amounts of annotated data, to investigate its potential in leveraging unannotated data for improving ASC performance. Our results demonstrate the efficacy of FixMatch in semi-supervised settings and highlight potential future directions based on a detailed error analysis.