A Multiscale Resonant Spiking Neural Network for Music Classification

Recent years have witnessed a boom of massive musical creations, demanding efficient classification models to better organize and utilize them. With mobile devices becoming the dominant approaches to music, the light weight and mobile deployabiliy of music classification models are also of growing importance. Artificial Neural Networks(ANNs) have been the mainstream paradigms for music classification, but problems concerning with computational and structural complexity hinder their further application to mobile devices. The Brain-inspired Spiking Neural Networks(SNNs) can process temporal information in a computationally economic way, which provides a possibility for mobile processing of music. To make a paradigm for music classification, we proposed the Multiscale Resonance SNN model that can comprehensively utilize the rich musical temporal information. With only binary activated neurons and sparse information flows, our model have achieved comparable music classification performance in various datasets.