The performance of automatic speaker identification systems degrade drastically in the presence of noise and other distortions, especially when there is a noise level mismatch between the training and testing environments. In this experimental research we have studied a recently robust front-end algorithm based on Gammatone Frequency Cepstral Coefficients GFCC combined to Cepstral Mean Normalization CMN technique. Our system using a Gaussian Mixture Models GMM classifier are implemented and tested under MATLAB®7 programming environment with multilevel White Gaussian Noise WGN applied to a test utterances using our proper database containing 51 Arabic speakers. Our aim is to study the performance of this suggested architecture and make a comparison with the conventional Mel Frequency Cepstral Coefficients MFCC method which we have successfully implemented and tested in the previous work. The obtained experimental results confirm the superior performance of the proposed method over MFCC and outperform it in different noisy environments. Our evaluations based on the recognition rate accuracy show that both MFCC and the proposed feature extractor have perfects performances in low-noise environments when Signal per Noise Ratio SNR is greater than 35 dB (practically 100% in all cases), but when the SNR of test signal changed from 0 to 40 dB, the average accuracy of the MFCCs methods is only 50.05%, while the proposed GFCCs extractors combined to CMN normalization still achieves an average accuracy of 55.43%.