A Novel Approach for Selection of Features for North Indian Classical Raga Recognition of Instrumental Music

摘要

Music information retrieval is a worldwide current research area. Lot of work has been done in Western music and now many researchers have focused their attention towards Indian music. Indian music is typically categorized as North Indian classical music and South Indian classical music. This paper proposes a novel approach to feature selection for recognition of North Indian classical music. The results were examined by multi-SVM (Support Vector Machine) classifier. Features were extracted for 1200 audio files of six ragas, viz Pooriya Kalyan, Yaman, Bihag, Bageshri, Malkauns, and Lalit. All these ragas were played on musical instruments such as Sitar, Sarod, Santoor, and Flute. The novelty of this work is that the database includes all style of music files having low or high rhythm and in any scale. Due to modification in feature set; accuracy for some ragas was improved up to 92% but an overall improvement was observed for all ragas.