One solution of extension of mel-frequency cepstral coefficients feature vector for automatic speaker recognition

Abstract

One extension of mel-frequency cepstral feature vector for automatic speaker recognition is considered in this paper. The starting feature vector consisted of 18 mel-frequency cepstral coefficients (MFCCs). The extension was done with two additional features derived from the appropriate spectral maximums of the speech signal. The main idea behind this research is that it is possible to increase the accuracy of automatic speaker recognition which uses only MFCCs by adding additional features based on the energy maximums in the appropriate frequency ranges of observed speech frames. In the experiments, accuracy and equal error rate (EER) are compared in the case when feature vectors contain only MFCCs and in cases when additional features are used. For the case of maximum recognition accuracy achieved (92.94%), recognition accuracy increased by around 2.43%. EER values have smaller differentiation, but the results show that adding proposed additional features produced a lower decision threshold. These results indicate that tracking of proposed spectral maxima in the spectrum of the speech signal leads to more accurate automatic speaker recognizer. Determining features which track real maxima in the speech spectrum will improve the procedure of automatic speaker recognition and enable avoiding complex models.