Machine learning based classification of violin and viola instrument sounds for the same notes

Abstract

In this paper, a study on the classification of violin and viola instrument sounds in the same notes that are hard for people to distinguish is presented. First, 16 statistical features defined in the time and frequency domain for 512 violin note recordings and 512 viola note recordings are extracted. Classification algorithms including Linear Distinction Analysis (LDA), k-nearest neighbors (k-NN), Support Vector Machines (SVM), and Random Forests (RF) classifiers are used to recognize violin and viola instrument sounds. The classification operation is performed on three different sets of features: only time domain features (12), only frequency domain features (4), and both time domain and frequency domain features (16). Simulation results show that the highest accuracy rate is obtained when using the Random Forests classifier with both time and frequency domain attributes. With Random Forest classifier, while obtaining 64.4% classification accuracy with 10 times cross-validity using only time domain and only frequency domain features, 79.6% classification success is achieved with both time and frequency domain features.