Speaker Identification plays an important role nowadays and gained major popularity. Voice-activated security systems are more efficient than other systems like biometric systems. Our paper addresses the need for natural speech and whispered speech in speaker identification. Even if the natural speech can be forged, the whispered speech cannot be forged easily, because the signal-to-noise ratio (SNR) is low. When whispering, there is no glottal stimulation or vibration in the vocal cords. We use different feature extraction techniques such as 2D Gabor filter-bank features, 1D Gabor filter-bank, MFCC (Mel frequency cepstral coefficients), and EMD based approach. We use these methods for examining both natural and whispered speech. These extracted features were classified using different classifiers including Decision Tree, Support Vector Machine (SVM), Fine K Nearest Neighbor (KNN), Weighted KNN and Random Forest. In our project we can observe EMD based features gives maximum accuracy of 100% for normal speech in Fine KNN, SVM, Decision Tree and Random Forest methods. Also, EMD based feature gives maximum accuracy of 93.75% for whispered speech in every classification method.