The Speaker identification is the process of determining which registered user provides a given utterance. In this paper, a powerful combination between the Discrete Wavelet Transform (DWT) and logarithmic Power Spectrum Density (PSD) is used for speaker first five formants extraction of one utterance, that are used as distinguishable classification features. As a classification method, the new approach by K-means algorithm is proposed, which uses the average of sums of point-to-centroid distances in the 1-by-K vector. To verify the experimental analysis for this work a Matlab simulation is performed and gave an excellent capability of features tracking even with 0dB SNR. This work is verified for text-dependant security systems applications such as password or PINs identification. Moreover, the attained results show excellent performance in classifications, which reaches about 94% classification rate.
