Classification of Environmental Sounds Using Spatio-temporal Based Spiking Neural Network

Environmental sound recognition (ESR) is an important task to ease everyday living by providing critical information on unpredictable events, and thus helps avoid dangers as well as prevents the occurrence of upcoming hazards. Despite advancements, machine learning and deep learning based frameworks often face challenges in efficiently processing temporal sequences and incur high computational costs in real-time scenarios. This study examines ESR classification where a spatio-temporal based strategy is implemented through spiking neural networks (SNNs), which is proven to be more efficient and accurate as compared to conventional networks, along with a superior utilization of satisfactory resources. The method is evaluated on a prominent publicly available dataset, achieving 99.38% accuracy with a notable improvement over state-of-the-art techniques. This highlights the potential of SNN to greatly improve the field of ESR by showing improved learning of complex and chaotic patterns inherent in environmental sound signals in an interpretable way.