Audio Signal Processing and Musical Instrument Detection using Deep Learning Techniques

Abstract

The advance of deep learning and audio signal processing techniques has led to serious development on Musical Information retrieval (MIR). Effective audio processing can improve speed, reduce errors, and sometimes increase the accuracy of detecting musical instrument. Spectrographic data is also necessary for many mathematical tools common across Musical Information retrieval. A major aspect of MIR is the categorization of pieces of music. One of the main tools used for categorization tasks in recent years is deep learning, which has led to many advancements in MIR. One such categorization task that deep learning is useful for is the recognition of instruments in a piece of music. In this paper, a new architecture is proposed for audio processing and musical instrument detection using Multilayer Perceptron (MLPs), Convolution Neural Networks (CNN), and Recurrent Neural Networks - Long Short Term Memory (RNN-LSTM). In addition, a number of experiments are implemented using real dataset that contains 20,000 recording. The three deep learning techniques are implemented and compared to present potential new solutions. The usage of processing techniques unique to the field of deep learning is also discussed.