Nowadays, research on Arabic language for non-native speakers is becoming more essential due to the usage of Arabic language by all Muslims for instance, during recitation of the Holy Quran. Hence, this research explores the development of speech processing as to ensure accuracy and efficiency is attainable since these elements are the most vital part in a speech recognition system. As we know, during recording process, noises will be recorded with the voiced signal and thus these noises need to be removed in order to produce robust speech recognition system. Thus, this study proposed to evaluate the effectiveness of Multiscale Principal Component Analysis (MPCA) along with Zero Crossing Rate (ZCR) as noise removal. Next Mel-Frequency Cepstral Coefficients (MFCC) is used as feature extraction whilst Dynamic Time Warping (DTW) approach for recognition process. Initial results proven the effectiveness of MPCA and ZCR based on the recognition rate attained for Arabic letters corpus uttered by Malay speakers as database.