An Efficient Visual-Based Method for Classifying Instrumental Audio using Deep Learning
Justin Hall, Wesley O'Quinn, Rami J. Haddad · 2019
In this paper, an efficient method for classifying and identifying instrumental audio is proposed via utilizing a deep learning image classification algorithm. The method of classification will involve analyzing the visual equivalent of an audio sample with a neural network to identify the generating musical instrument. Audio samples are converted into a logarithmic spectrogram format, which allows visual classifiers to attempt the identification of the audio source. The primary focus is on developing an efficient method for analyzing audio spectrograms using various forms of neural networks and analysis techniques. The use of deep learning convolutional neural networks in analyzing visually formatted audio data provides an enhanced classification method over traditional schemes. A classification accuracy of 73.7% was achieved with a limited data set and minimal manipulation of network architecture.