Paper Title
ARABIC SPEAKER IDENTIFICATION BASED ON PHONEMES

Abstract
In the past few decades, there has been a rapid development in the field of speaker identification. Speaker identification is an important application of digital speech processing and machine learning in user authentication. There are still challenges degrading the reliability of the speaker identification systems. These challenges include talking in a noisy environment and variation in the emotional conditions of speakers while talking such as surprise, anger, happiness, or sadness and length of the voice. Even though the performance of the speaker identification systems has improved with the current research and developments in the speech processing and machine learning methods, an enhancement is still required with respect to text-independent speaker identification in Arabic language. Therefore, this research aims to propose a reliable system architecture for Arabic speaker identification to identify speakers based on short voice input such as phonemes. The system exploits unique features of the speech signal for the purpose of speaker identification. Features extraction is an important step for this task and a number of features such as mel-frequency cepstral coefficients (MFCC) and delta mel-frequency cepstral coefficients (ΔMFCC) are employed. In addition, various machine learning algorithms such as Gaussian Mixture Model (GMM), Support Vector Machine (SVM), and Gaussian Mixture Model-Universal Background Model (GMM-UBM) have been used for training models for identification process. The results show that the GMM model performs better in identifying Arabic speakers based on single phoneme with an accuracy of 78.30% and SVM performs better for double phonemes with an accuracy of 89.42%. Keywords - Speaker identification, Arabic, Phonemes, MFCC, Machine Learning, GMM, SVM, GMM-UBM.