Spectral Masking in MFCC Calculation for Noisy Speech
K. K. Tomchuk · 2018
To increase noise immunity of the MFCC (mel-frequency cepstral coefficients) widely used for voice signal parametrisation it is suggested to use a psychoacoustic model of frequency masking. Additionally, taking into account the formation mechanism of the formant regions in the voice signal spectrum it is suggested to influence the spectral samples corresponding to multiple harmonics of the fundamental tone. The modified algorithm is investigated on the basis of the single word recognition system adapted for MFCC voice signal parametrisation only. The positive effect of using proposed additional voice signal transformation in the parametrisation algorithm is shown.