Speech recognition method.

摘要

An auto-correlation analyzing unit (4) calculates an auto-correlation coefficient vector based on a digital signal of a speech, and a linear predicting unit (5) calculates a linear predictive coefficient vector from the auto-correlation coefficient vector. A cepstrum analyzing unit (6) calculates a cepstrum coefficient vector from the linear predictive coefficient vector, a dynamic cepstrum generating unit (7) provides a time frequency masking lifter on the cepstrum time sequence, and generates a time sequence of the dynamic cepstrum vector. When a switch (SW1) is switched, a centroid generating unit (9) carries out vector quantization of the dynamic cepstrum, the speech is represented by a sequence of vector codes by a vector quantizing unit (10), and the sequence of vector code is recognized by a HMM recognizing unit (13) based on the HMM obtained by HMM learning. <IMAGE>