Melodic pattern recognition in Indian classical music for raga identification

Abstract

Melody is the soul of the Indian classical music for raga identification. This paper evaluates different algorithms proposed for raga recognition for effectiveness and computational cost. As per the analysis done, the pitch class distribution and N-gram approaches found out to be more effective. It was revealed that none of the research focused on the facet of minimum duration sample required for identification. The main aim of the experiments is to identify the least duration sample required for identification of raga. Least duration will lead to less computational cost and time. Dataset used is voiced audio samples of monophonic music with duration of audio samples ranging from 30 to 180 s from the beginning of raga rendition. Pitch extraction for melodic data is done using auto correlation method in tool praat. Findings revealed that different ragas require varied duration for accurate identification. Potential directions to improve the raga identification performance with less possible duration are proposed.