Speaker Recognition Based on Weighted Mel-cepstrum

The key point of speaker recognition is to extract the unique, effective, stable and reliable features that can stand for the personality of the speaker from the speech signal. This paper applied the psychologically weighted technology in mel-cepstrum analysis and adopted the Signal-to-Mask Ratios (SMRS) obtained from the psychoacoustic model as the weighting function to obtain the weighted mel-cepstrum coefficients (WMCEP) as features in speaker recognition. Experiments showed that the WMCEP not only described the speaker´s formant much better than MFCC and MCEP, but also had robustness to some extent for speaker recognition.