| 摘要 |
PROBLEM TO BE SOLVED: To provide a speech recognition device having higher recognition performance than hitherto without increasing a recognition processing time, by varying an analytical window length according to the characteristic of an input speech, by utilizing the fact that the input voice has both a part having high time constancy, such as a vowel and a part having low time constancy, such as a consonant. SOLUTION: Speech analytical parts 103, 104 convert an inputted voice from a speech to each characteristic vector with each different analytical window length at each time. An inter-characteristic vector correlation calculation part 105 calculates a correlation value between each characteristic vector at each time. The optimum characteristic vector is selected from among characteristic vectors calculated with different analytical window lengths corresponding to correlation values between the characteristic vectors, and the voice is recognized by an HMM method by using it.
|
