A computational approach for identifying assamese folk music instruments

Abstract

The classification of musical instruments by using a computational technique is a very challenging task. The developments in signal-processing and data-mining techniques have made it feasible to analyse the many musical signal characteristics, which is essential for resolving the classification issues in music. In this work, 12 popular Assamese folk music instruments were selected for identification. Twelve musicians played the instruments and audio samples were recorded, different instantaneous features were extracted, and an effort has been made to identify those instruments using three popular classification techniques - Decision Tree Classifier (DTC), Support Vector Machine (SVM), and Linear Discriminant Analysis (LDA). A performance-based comparison was made among the three classifiers. The proposed sets of features enabled the DTC, SVM and LDA models to achieve average accuracy ratings of 86.9%, 90% and 92.2% respectively. Regarding the performances of the three fitted models in identifying instrumental sounds, this study offers a valid comparison.

Keywords

• musical instrument
• machine learning
• decision tree classifier
• support vector machine
• linear discriminant analysis