Classification of male and female speech using perceptual features

Abstract

Gender identification systems nowadays, are gaining momentum in terms of popularity because of their wide areas of application. They can be used in a variety of fields ranging from security and authentication services to content based information retrieval and also criminal investigations. Gender detection has started to gain importance because of the fact that recent studies conducted showed that the performance of gender dependent speech recognition models performs much better than gender independent models. In the proposed work, we aim to build such a system involving perceptual audio features such as pitch and tempo based features, short time energy etc., which are used to train classifiers to differentiate between the two classes of gender. We have selected such a combination of features as because previous works focused only on either pitch approach, MFCC approach etc., whereas our work is perhaps one of the first involving a combination of several such perceptual features. The system was tested on a wide range of speech files and was shown to be yielding promising results.