| 摘要 |
PURPOSE: To improve the recognition rate of an inputted voice by extending/ shortening the time length of a registering pattern in accordance with unknown input pattern length. CONSTITUTION: The time length of a registering pattern is extended/ shortened in accordance with unknown input pattern length. Namely a time length rate calculation part 6 finds out the rate (αj) of time length between a registering pattern having the largest degree of similarity and an input voice pattern based upon the collation result of a collation part 3. A time length extending/ shortening part 5 extends/shortens the time length of all registering patterns based upon the time length rate (αj) to generate a word pattern dictionary 7, which is used for collation at the time of succeeding pattern matching. The time length is extended/shortened based upon a follow-up degree preventing the sudden extension/shortening of the time length so as not to be influenced by misvocalization or the variation of time length of individual vocalization. Consequently, the reduction of the recognition rate is efficiently prevented.
|
