Performance Comparison of CNN and AST Models with Stacked Features for Environmental Sound Classification

A new study shows that Convolutional Neural Networks (CNNs) using stacked audio features can match or outperform larger Audio Spectrogram Transformer (AST) models for environmental sound classification when data or compute is limited. Tested on ESC-50 and UrbanSound8K datasets, these CNNs offer a more computationally and data-efficient path, making them ideal for resource-constrained applications like smart city monitoring, acoustic surveillance, and edge-level quality control without needing massive pre-training.