Music Instrument Recognition from Spectrogram Images using Convolution Neural Network

Abstract

The projected system presents a unique approach to instrument Recognition (MIR) supported Convolution Neural Networks (CNNs). Previous MIR strategies are supported planning and extracting spectrogram features from the audio signal so as to explain its characteristics and classify it. In distinction, CNNs learn the features directly from the input file. The model has evidenced successful in solving various advanced multimedia system information Retrieval (MIR) issues, like image classification or voice recognition. The projected system seeks to explore whether or not the success may be imported to MIR also. The results from this work shows 97% accuracy for all instrument classes.