Feature selection and classification of indian musical string instruments using svm

Abstract

Music is one of the most important aspects of human life. It touches your soul and relaxes your mind and body. Many musicians from all over the world share their music on social media platforms. On the internet, you can listen to a wide range of music. It is difficult and time-consuming to search for a specific one based on the singer, genre, music piece, instrument, and so on. Previous attempts were mostly concerned with categorising various instruments belonging to different families, such as woodwind, brass, string, percussion, and so on. The objective of this paper is to categorise musical instruments of the string family, particularly Indian musical instruments of the strucked and plucked types. Experimentation is carried out using monophonic recordings of solo instrument artists. Temporal, spectral, statistical, and the first 13 Mel-frequency cepstral coefficients have all been considered in audio features. Support Vector Machine (SVM) with linear, radial basis, and quadratic kernels was used for classification. The first two experiments filter features using MANOVA and the Chi-square method. In the third experiment, temporal, spectral, statistical, and MFCC features are chosen for classification independently. The highest accuracy, 93.4 percent, was obtained by using the top five selected features. For selected features, there is only a +1% variation in accuracy when using SVM with different kernels. For MFCC 13 features, the linear kernel had an accuracy of 92.5 percent, while the radial basis kernel had an accuracy of 52.6 percent.