A new approach for gender detection from voice data: Feature selection with optimization methods

Abstract

In recent years, gender detection has been an important problem in speech analysis, with many different applications. Different features of sound data such as pitch, median, and frequency are used for gender detection. In this study, a feature selection method based on metaheuristic optimization algorithms is proposed for gender detection from voice data. In the proposed method, the feature set representing the voice data in the most appropriate way is selected with optimization algorithms, and gender detection has been implemented with artificial intelligence algorithms using the obtained features. Nature-inspired metaheuristic optimization algorithms, which are capable of solving complex problems, are used to select features from voice data. Particle Swarm Optimization (PSO), Ant Colony Optimization (ACO), Salp Swarm Algorithm (SSA), and Whale Optimization Algorithm (WOA) have been modeled for the first time for feature selection from voice data. A publicly accessible data set has been used to measure the efficiency of metaheuristic optimization algorithms. The performances of PSO, ACO, SSA, and WOA for feature selection have been compared in terms of three different criteria: fitness function value, accuracy value, and the number of selected features. After feature selection with metaheuristic optimization algorithms, Naive Bayes and Decision Tree algorithms have been applied to the new data sets obtained and the original data set. As a result of the analysis, it was observed that this method which uses metaheuristic optimization algorithms for feature selection increased the success rate in the results obtained with Naive Bayes and Decision Tree algorithms.