Raaga Recognition System using Gaussian Mixture Model for Gender Dependence and Independence: Melakartha Raagas

Main Article Content

Tarakeswara Rao B
Dr. Prasad P.V.G.D, Reddy

Abstract

Recognizing Melakartha raagas from speech has gained immense attention recently. With the increasing demand for human
computer interaction, it is necessary to understand the state of the singer. In this paper an attempt is made to recognize and classify the raagas
from the singers database where the classification is mainly based on extracting several key features like Mel Frequency Cepstral Coefficients
(MFCCs) from the speech signals of those persons by using the process of feature extraction. For training and testing of the method, data is
collected from the existing database (both genders) with due verification relating to melakartha raagas of singers from Andhra Pradesh, India.
Out of 72 melakartha raagas (36 male 36 female) available for training, a few raagas (both genders) were specifically selected and tested. Then it
is found that all the tested raagas are well recognized. In another case the 50 melakartha raagas for training and another 22 raagas for testing. The
experiments were performed pertaining to singer dependent and independent raagas. Using a statistical model like Gaussian Mixture Model
classifier (GMM) and features extracted from these speech signals, an unique identity was built for each raaga that enrolled for raaga
recognition. Expectation and Maximization (EM) algorithm, an elegant and powerful method, is used with latent variables for finding the
maximum likelihood solution, to test the other raagas against the database of all singers who enrolled in the database.

 

Keywords: Raaga Recognition, Gaussian Mixture Model (GMM) classifier, Sequential Forward Selection, EM algorithm, Mel Frequency
Cepstral Coefficients (MFCCs), Gender Dependence and Independence.

Downloads

Download data is not yet available.

Article Details

Section
Articles