Gender Identification from Speech Recognition Using Machine Learning Techniques and Convolutional Neural Networks

Abstract

Gender identification represents a fundamental component of speech recognition and automatic interacting sound responding systems. Identifying the voice gender minimizes the computational loads of these systems for additional processing. Standard approaches for gender estimation from the speech have broadly relied on the extraction of speech features and classification tasks. This paper proposes a technique for gender identification of speech samples using the speech recognition process. The proposed technique extracts essential voice features like Mean, Zero-Crossing, Standard Deviation, and Amplitude, as well as 12 most significant features from every voice sample, and combines them to create voice feature vectors. The proposed technique uses several machine and deep learning methods such as Random Forest, KNN, Logistic Regression, Decision Tree, and CNNs, in order to classify the voice vectors into Male and Female classes. After comparing the evaluation metrics results of all classifiers, the proposed technique finds out that the CNN model is the best classifier used to classify the voice vectors with a higher precision value of 1.0.