IN SEARCH OF AN EFFICIENT 'CHANDAS' IDENTIFICATION SYSTEM FOR MALAYALAM POEMS FROM AUDIO

The music genres in Kerala are characterized by a syllabic level rhythmic pattern called ‘Chandas’. The ‘Chandas’ basically is a syllabic level rhythmic pattern characterized by some rules for arranging light and heavy syllables in a poem or a song. This paper explains the two different approaches for identifying this rhythmic pattern of traditional Malayalam poems or songs from known music genres in Kerala. The first method obtains ‘Chandas’ by counting the number of syllables per two or four lines of a poem. The major step in this process includes detection of syllable nuclei in the audio signal. For this purpose intensity contours are used. The second method is an ANN based machine learning approach which can incorporate most statistical characteristics of these poems or songs.