Abstract

Human voice carries an abundance of paralinguistic information, making it an invaluable resource for various voice recognition applications. Gender classification is a crucial aspect of voice signal analysis, albeit a challenging task. To distinguish between male and female voices, a range of techniques is applied to identify pertinent features for model development using training data. These models play a key role in determining the gender of a given voice signal. This research offers a comprehensive analysis of well-established voice signal features using a prominent dataset. It explores various machine learning models from different theoretical families for gender classification and employs three feature selection algorithms to identify optimal features that enhance model efficiency. Experimental results emphasize the significance of specific subfeatures in boosting classification model performance. Notably, the best recall value achieved was 99.97% without feature selection, 99.7% for two deep learning (DL) and support vector machine (SVM) models, and 100% for SVM techniques with feature selection

Description