Voice intensity based gender classification by using Simpson's rule with SVM

The proposed technique measures the voice intensity of an utterance by calculating area under the curve. The curve is obtained by normalizing the cubic polynomial fitted through the peaks. These peaks are found from each frame of the utterance when it is divided into segments of 20 milliseconds. The Simpson's rule is used to calculate area under the curve and SVM uses this area to classify the genders. The use of one dimensional feature, area of utterance, is an evidence for the time and computational efficiency of this technique. The aspects observed in this paper, for the validity of the technique, are: it works for different natural languages, independent of recording equipment, any text can be used for the classification, and its biasness when different number of male and female speakers are used for the training of the system. A promising classification rate of 98.27% is achieved.