Estimation of Recall Values and Accuracy of Gender Identification for the Different Age Groups Based on Voice Signals

Abstract

Speech is an crucial channel for exchanging ideas between human beings. The coordinated movement of the muscular part of vocal tract is main reason for generating the voice signals. So, voice signals can carry the essential characteristics of the human beings as: origin, emotions, age, ascent of language, gender, health status, etc. Sex of the human being can be recognized based on the analysis of the voice signals. This present paper mainly emphasises the gender-wise recall values, i.e., male and female, and identification accuracy of the gender for the proposed model based on several classes of the age groups of the speakers as Old, Teens, and Middle. To compute the gender-wise recall values and accuracy of gender identification for the proposed model, Mel frequency Cepstral Coefficients (MFCCs) are utilized as a parameter of the speech signals. A machine learning algorithm is used for the classification of gender. Recurrent Neural Network Bi-directional Long Short-Term Memory (RNN-BiLSTM) is used as a machine learning algorithm with the adaptive Movement estimation (ADAM) optimization technique. Recall values for the speakers that are belongs to female as well as related to old age category, are the maximum in compare to others categories. In the concern about recall values for the speakers that are related to the male, the teenage category has the maximum. The highest identification accuracy of the gender for the proposed model is computed as 92.20%. This value relates the old age category. After the computation of the recall values in the all-age groups, it is observed that the middle age group shows the worst performance for male as well as female. It is also observed that the identification accuracy of gender for middle age group shows the lowest values.