Gender Classification Based on Voice Using MFCC Features and Naïve Bayes Algorithm

This study aims to classify gender based on voice signals using Mel-Frequency Cepstral Coefficients (MFCC) and the Gaussian Naïve Bayes algorithm. Voice-based gender classification is relevant to biometric identification, speech-based services, and human–computer interaction. This study uses 150 audio recordings from 150 different speakers, consisting of 75 male and 75 female speakers. The audio was converted to mono, resampled to 16 kHz, peak normalized, and extracted into 13 MFCC coefficients. The mean MFCC values were used as input features for the Gaussian Naïve Bayes classifier. Four training and testing data split scenarios were evaluated, namely 50:50, 60:40, 70:30, and 80:20. The results showed accuracy values of 88.00%, 96.67%, 95.56%, and 96.67%, respectively. The highest accuracy of 96.67% was obtained using the 80:20 split, with 29 of 30 testing samples correctly classified. These results indicate that the combination of 13 MFCC coefficients and Gaussian Naïve Bayes can effectively classify gender based on voice signals. The study concludes that the proportion of training data affects classification performance, while further research using larger and more diverse datasets is needed to improve model robustness and generalizability.