Deep Neural Network Based Poetic Meter Classification Using Musical Texture Feature Fusion

Abstract

In this paper, a meter classification scheme is proposed using musical texture features (MTF) with a deep neural network (DNN) and a hybrid Gaussian mixture model-deep neural network(GMM-DNN) framework. The performance of the proposed system is evaluated using a newly created poetic corpus in Malayalam, one of the prominent languages in India and compared the performance with support vector machine (SVM) classifier. Initially, a baseline-mel-frequency cepstral coefficient (MFCC) based experiment is performed. Later, the MTF are fused with MFCC. Whilst the MFCC system reports an overall accuracy of 78.33%, the fused system reports an accuracy of 86.66% in the hybrid GMM-DNN framework. The overall accuracies obtained for DNN and GMM-DNN are 85.83%, and 86.66%, respectively. The architectural choice of DNN based classifier using GMM derived features on the feature fusion paradigm showed improvement in the performance. The proposed system shows the promise of deep learning methodologies and the effectiveness of MTF in recognizing meters from recited poems.