Pengembangan model probabilistic neural network (PNN) pada pengenalan kisaran usia dan jenis kelamin berbasis suara

Abstract

Voice signal can be used to represent a speaker. Voice can also be used to recognize the age range and the gender of the speaker, because of the difference in frequency of the voice signal for different age and gender. Age range for this experiment is divided into children (8-10 years old), teenagers (12-17 years old) and adults (30-50 years old). We performed feature extraction to obtain a representation of the speaker using Mel-Frequency Cepstrum Coefficient (MFCC). MFCC is used with 13, 20, 26 coefficients. Probabilistic Neural Network (PNN) is used as the feature matching model. The PNN model is constructed from three different proportions of TRD (Training data) : TSD (Testing data) (25%:75%, 50%:50%, 75%:25%). MFCC with 26 coefficient gives the best accuracy of 93.47%. The proportion of 75% (TRD) : 25% (TSD) also gives the best accuracy of 94.22%. PNN-age and PNN-gender identification result in an accuracy of 91.26%. Specifically, PNN accuracy percentage for male and female are 92.26%, and 95.20%, respectively, while for children, teenagers and adults the accuracies are 99.85%, 92.28% and 95.57%, respectively. In conclusion, the average of accuracy for all categories is 91.26%. The experiment using the voice of male-teenagers gives the worst accuracy, which is caused by the overlapping frequencies of voice in children, teenagers and adults.