GA Algorithm Optimizing SVM Multi-Class Kernel Parameters Applied in Arabic Speech Recognition

Abstract

Objectives: This paper proposes a novel recognition technique (ASR) based on GA optimized SVM multi-class algorithm. Methods/Statistical Analysis: The Kernel parameters of support vector machine are very important problems that have a great influence on the performance of recognition rate. Thus, GA is adapted to optimize the penalty parameter C and the kernel parameter λ for SVM multi-class, which leads to improve classification performance. Finally, the proposed model is tested experimentally using eleven Arabic words mono-locator. Each word of them is extracted by Mel Frequency Cepstral Coefficients (MFCCs) and used as an input to the SVM multi-class classifier. Findings: The proposed method enhances the recognition rate which is performed to 100% within short duration training time. Application/Improvements: The obtained results shows that the GA-SVM technique achieved the better performance