Speaker Recognition Using Principal Component

摘要

This paper propose a new feature vector-Mel Frequency Principal Coefficient (MFPC), applied to speaker recognition. It is derived by performing Principal Components Analysis on the Mel Scale Spectrum Vector. Compared with conventional Mel Frequency Cepstrum Coefficient. MFPC efficiently exploited the correlation information among different frequency channels. These correlations, which is mainly caused by the vocal tract resonance, have been found to vary consistently from one speaker to another. And we select these feature coefficients according to their Fisher Ratio, which will guarantee the largest discriminability between classes in the given dimensionality. Finally, we implement a text-independent speaker recognition system. It uses Vector Quantization to design codebooks of given reference speakers. The experiment results demonstrate that our proposed feature vector has characteristics of compactness, large discriminability and low redundancy.