| 摘要 |
PURPOSE:To establish a recognition candidate only when a difference in continual period of time is realistic by predicting the continual period of time of a recognition unit to be recognized next by a continuance prediction part and performing collation by a collation part by utilizing the predicted continual period of time. CONSTITUTION:A speech section detection part 3 detects a speech section by an analytic part by using an LPC cepstrum 0th-order coefficient. Then a next-syllable candidate prediction part 4 selects syllables to be recognized next, at every frame, by using a grammatical rule. Then a continual period of time prediction part 5 calculates the continual period of time of a next syllable candidate. Namely, mean continual period of time at every kind of syllable is store in a standard speech kind classified mean continual period of time storage buffer 13, and the continual period of time of the input speech part of the input speech which are already recognized is stored in an input speech continual period of time storage buffer 11; and the continual period of time prediction part 5 predicts continual period of time at every recognition unit of the input speech to be recognized by using the rate of continual period of time of the syllables which are already recognized to the mean continual period of time and the collation part 6 performs the collation within a certain range centering on the predicted continual period of time. |
