Correlative consideration concerning feature extraction techniques for speech recognition — A review

摘要

This paper frames co-relation on three feature extraction techniques in ASR system. As compared to primarily used technique called MFCC (Mel Frequency Cepstral Coefficients), PNCC (Power Normalized Cepstral Coefficients) obtains impressive advancement in noisy speech recognition due of its inhibition in high frequency spectrum for human voice. The techniques differ in the way as MFCC uses traditional log nonlinearity and PNCC processing substitute the usage of power-law nonlinearity. Experimental results relay on the fact that PNCC processing provides substantial improvements in recognition accuracy compared to MFCC as well as PLP (Perceptual Linear Prediction) processing for speech recognition in the existence of various types of additive noise and reverberant environments with marginally greater computational cost and the with the usage of clean speech, it does not lowers the decoding accuracy.