PROJECT TITLE :

Spectral–Temporal Receptive Field-Based Descriptors and Hierarchical Cascade Deep Belief Network for Guitar Playing Technique Classification

ABSTRACT:

In the field of audio signal processing, one area of particular interest is the retrieval of musical information. However, the playing techniques of musical instruments have received only a moderate amount of attention over the course of history. This study puts forward the idea of an algorithmic classification system for different styles of guitar playing (GPTs). Automatic classification of GPTs is difficult to achieve because different playing techniques only differ from one another in subtle ways. This work presents a new framework for GPT classification. To extract features from guitar sounds, it uses a new feature extraction method that is based on spectral–temporal receptive fields (STRFs). The classification of GPTs is accomplished through the use of a supervised deep learning approach in this work. For the purpose of carrying out automatic GPT classification, a fresh deep learning model has been proposed. This model goes by the name of the hierarchical cascade deep belief network (HCDBN). A number of simulations were carried out, and the following datasets were utilized to evaluate the effectiveness of the algorithms: 1) data on the beginnings of signals; 2) complete audio signals; and 3) audio signals captured from an actual environment. In setup 1, the proposed system achieves an F-score that is approximately 11.47% higher than the previous one, and in setup 2, it achieves an F-score that is 96.82% higher. The findings of setup 3 illustrate that the proposed system performs admirably in a setting that more closely resembles the real world. These findings demonstrate that the system that was proposed is reliable and has a very high level of accuracy in automatically classifying GPT data.