Classification of Stop Consonants using Modulation Spectrogram-Based Features

Abstract

In this paper, we propose use of modulation spectrogram-based features for stop consonants classification based on their place of articulation. Stop sounds are classified as bilabial, alveolar and velar according to their place of articulation. The modulation spectrogram which is a two- dimensional (i.e., 2-D) feature represents modulation of low frequency components with acoustic frequency. In this work, modulation spectrogram has been obtained for all stop consonants from TIMIT database and then a dimension reduction algorithm, viz., higher order singular value decomposition (HOSVD) is applied on the feature vectors. The reduced dimension feature set is then applied to a Support Vector Machine (SVM) classifier which gives an overall accuracy of 94.25% for stop classification and 95.29% for place of articulation classification.