Enhancing Human Voice Gender Classification: A Comparative Analysis of Support Vector Machine and Convolutional Neural Network Algorithms

To implement an efficient prediction model for human voice recognition, distinguishing between male and female using CNN and SVM algorithms, with enhanced accuracy, we propose a method leveraging feature selection through the Novel SVM algorithm in conjunction with CNN. The SVM, based on Recursive Feature Elimination, significantly improved the classification accuracy of the CNN on the dataset. Experimental results show that the CNN, after applying the Novel SVM, outperformed all other comparative algorithms, proving it to be highly effective for gender voice recognition. The study utilized a voice dataset with 56 samples and G-power values of 80%, containing 21 attributes, collected from Kaggle. The Novel SVM achieved an accuracy of 98.63%, while the CNN achieved 98.21%. A statistically significant difference was observed between the two algorithms $(p$ = 0.024; p < 0.05) with a 95% confidence interval. This study concludes that the Novel SVM algorithm is significantly more effective for predicting human voice recognition compared to the CNN algorithm.