Exploring Neural Networks for Musical Instrument Identification in Polyphonic Audio

The purpose of this article is to introduce neural network-based methods that surpass state-of-the-art models, either by training faster or having simpler architecture, while maintaining comparable effectiveness in musical instrument identification in polyphonic music. Several approaches are presented, including two authors’ proposals, i.e., spiking neural networks (SNNs) and a modular deep learning model named fully modular convolutional neural network (FMCNN). First, a CNN and a convolutional recurrent neural network (CRNN), adapted from literature, are built to detect up to 13 different instruments in polyphonic music. Furthermore, FMCNNs and SNNs are explored. The results obtained demonstrate that both FMCNNs and SNNs outperform traditional CNNs and CRNNs in terms of accurate instrument identification. Moreover, the SNN architecture is much less complex compared to other model sizes. These findings highlight the efficacy of the methods proposed in musical instrument identification in polyphonic audio.