| 摘要 |
<P>PROBLEM TO BE SOLVED: To provide a speech recognition device and a speech recognition method capable of effectively expressing characteristics of phonemes and vocabularies, and improving a recognition rate of speech including the phonemes and the vocabularies. Ž<P>SOLUTION: This speech recognition device 1 includes: a smoothing means 4a for determining a linear prediction coefficient by smoothing a speech signal at each frame; an autocorrelation function calculating means 4b for calculating an autocorrelation function of the smoothed speech signal; an utterance period probability calculating means 4c for calculating an utterance period probability at each frame on the basis of the liner prediction coefficient and the autocorrelation function; a noise ratio probability calculating means 4d for calculating a noise ratio probability at each frame on the basis of the autocorrelation function; and a speech recognition means 4e for performing speech recognition by calculating a gross probability on the basis of the utterance period probability and the noise ratio probability in addition to a state transition probability and an output probability by using a concealed Markov model. Ž<P>COPYRIGHT: (C)2010,JPO&INPIT Ž
|
