ID FEATURES FOR SPEECH RECOGNIT1
Kuldip K. Paliwal · 1998
Cepstral coefficients (LP) analysis or from monly used features in tion systems. In this centroids as new features cepstral features for features have properties they are quite robust to ported in the paper ju: tures as supplementary ~ derived either through linear prediction filter bank are perhaps the most comcurrently available speech recognipaper, we propose spectral subband and use them as supplement to spc'ech recognition. We show that these similar to formant frequencies and noise. Recognition results are retifying the usefulness of these feafeatures.