A Text Independent Speaker Recognition System Using Gaussian Mixtures Model, LPC-Cepstral, and Pitch Information

In the robust speaker recognition system, the use of spectrum features, especially the LPC-Cepstral coefficients, has been successful because their computation is easy and used together with the Gaussian Mixture Model. GMM have shown to provide a good performance in speaker recognition problems. However, when the LPC-Cepstral coefficients are calculated from the parameters obtained from LPC analysis, useful information of the speaker is still ignored. To improve the performance of the GMM based speaker recognition systems, this paper proposes to use a speaker feature vector derived from a combination of LPC-Cepstral coefficients and the pitch period. This allows increasing the recognition rate without degradation of the system robustness. Computer evaluation results show that the proposed system achieves a 98% recognition rate, even in situations where the recognition rate of other previously proposed GMM based recognition systems was only slightly higher than 85%.