Hybrid Features Extraction and Machine Learning Based Arabic Speaker Classification

Abstract

In this era of technological advancement the machine learning and the artificial intelligence are becoming vital techniques which are extensively employedfor the establishment of smart cognitive systems. In this context, a hybrid model based Arabic speaker recognition tactic is devised. The target is to attain an effectual way out with an elevated accuracy. It is reachable by tactfullyutilizing the hybrid features extraction and the robust classification tactics. The incoming Arabic speech is denoisedand conditioned by using appropriate pre-conditioning. The Perceptive Linear Prediction Coding Coefficients (PLPCC) and the Mel-Frequency Cepstral Coefficients (MFCCs) are mined from the enhanced Arabic speech. Afterward,k-Nearest Neighbor (KNN) classifier is employed to recognize the speaker. The system attains an Arabic speaker classification accuracy of 90.8 %. It confirms the interest of embedding the designed framework in modern systems such as smart spaces like buildings, offices and homes for an effective realization of resources sharing and management.