Perbandingan Metode Pengenalan Pola Suara Menggunakan Codebook dan Probabilistic Neural Network Berdasarkan Kisaran Usia dan Jenis Kelamin

Abstract

Voice signal be used to identify a speaker, including the age range and gender based on the difference of its frequency characteristic. This research compares two method of voice identification namely codebook and probability neural network (PNN) in recognizing the age range and gender of the speaker. In this research, the age range is divided into three categories namely children (8-11 years old), teenagers (12-17 years old) and adults (30-50 years old). Each age category is divided based on gender, so that there are six categories in total. This research utilized 600 voice data representing the total six categories. MFCC is used as a method of feature extraction, whereas K-means is used as the clustering method. Several important parameters in the MFCC process are the number of cepstral coefficients, overlap, and time frame. The overlap and time frame values are 0.5 and 40 ms, respectively; whereas the chosen cepstral coefficients to produce the maximum accuracy are 13, 20, and 26. The comparison of voice identification is constructed from three different proportions of training data and testing data (25%:75%, 50%:50%, 75%:25%). It is shown that the accuracy of codebook method is 97.20% whereas that of PNN is 95.17%.