Elevating Environmental Sound Classification with Stacked Convolutional Neural Networks

Environmental sound classification (ESC) stands as a cornerstone in audio recognition, with far-reaching applications encompassing audio surveillance and forensic investigations. Yet, ESC presents formidable hurdles owing to the nuanced complexities inherent in environmental sound data, diverging significantly from conventional audio sources. Our study delves into ESC utilizing the UrbanSound8k dataset to investigate ESC using Stacked Convolutional Neural Networks (CNNs). Through meticulous experimentation, we probe the intricate interplay of hyperparameters, including layer count and filter configurations, to finely tune ESC accuracy. Notably, our method yields a peak accuracy of ${9 3. 8 2 \%}$, showcasing a significant improvement over existing benchmarks. This achievement underscores the efficacy of our approach in pushing the boundaries of ESC research, with implications for advancing state-of-the-art audio recognition technologies.