A realization of an automatic speaker identification

This paper describes a method of realization of an automatic speaker identification. Speaker identification was performed within a speech database of 20 speakers, regardless of the text that they uttered, based on speech features in the form of mel-frequency coefficients, while the speaker modeling was performed using the appropriate multidimensional Gaussian distribution. The recognition results are given for the speech database of 20 speakers. The number of MFCCs, the type of window functions and their overlap, are varied in the experiments.