Fuzzy Logic Based Speech Recognition and Gender Classification

Abstract

An approach of recognizing a person based on the individual information present in speech signals is named as speaker recognition. Nowadays, gender classification is a challenging factor in the speaker recognition. Different genders have dissimilar frequency ranges and respective pitch values. Perceptually and biologically, pitch is proved as a good discriminator between male and female voice. More formally, gender classification is done based on the relevant parameters. In this paper, our works aim to classify the gender of the speaker by using the MATLAB Fuzzy Toolbox. Mamdani fuzzy interface system is able to represent the gender classification based on the input variables: frequency and pitch. By the behavior of the input variables on the fuzzy rule based expert system, the output is predicted as male, female, and children. The work also extends to make the fuzzy controller adaptive. The test results show the reliability of performance. The proposed method is build to improve the robustness of the gender classification. Simulation results for male, female, and child accuracy of COA are nearly equal to 0.15 µ 0.001, 0.452, and 0.751 µ 0.001, respectively.