Musical Instrument Classification Based on Machine Learning Algorithm

Abstract

Musical instrument classification from an audio file is a very interesting and important topic in machine learning. In this paper, we represent a method to classify a musical instrument from a single audio file of a specific instrument. We focus on classifying six musical instruments that are very popular for Indian subcontinent, basically used to folk songs. It is also helpful for music genre classifiers. A fairly small dataset contains 600 audio files from harmonium, flute, monochord (ektara), cylindrical wooden drum (dhol), tawala, and violin that are classified using MFCC and various types of classifier. MFCCs are based on signal disintegration with the help of a filter bank. The great things of MFCCs over spectrogram is that they try to model the way perceive like frequency. To classify musical instruments, we are used as the k-nearest neighbor and support vector machine classifier with RBF kernel which provides optimum classification ability. A very high accuracy is achieved (97%) on the test set of our generated dataset.