CD-HMM MODELING FOR RAGA IDENTIFICATION

摘要

In this paper raga identification using Continuous Density Hidden Markov Model(CD-HMM) is proposed. The input is taken from the film song database.Training and testing of Raga identification using CD-HMM is carried out. Here, MFCC (Mel frequency Cepstral coefficients) feature alone is taken as features. In modeling, training is done using Hidden Markov Models (HMM). HMM is a type of indirect model that use statistical variation are modeled in an effective manner. Here Continuous density HMM is used since discrete HMM uses index values. In second phase maximum likelihood decision is applied to examine the raga. Finally the identification of raga is rated and accuracy is compared for different mixture values. The main objective in identification system of raga is that it forms the basic idea for content recovery for instrumentation in hindustanic, carnatic, folk and movie songs.