| 摘要 |
PURPOSE: A method for recognizing voice is provided to prevent the apparatus for recognizing voice from performing useless operation due to a noise or a malfunction. CONSTITUTION: A method for recognizing voice includes following steps. At the first step, test data for training recognizing a word is provided. At the second step, a statistical output on the length of arbitrary word from the test data is extracted. At the third step, the characteristics of the word is extracted by using the test data, and a word model is made up for each word to be recognized. At the forth step, the characteristics of a noise model is extracted by using a noise model test word, and the noise word model is made up. At the fifth step, a probability of each word and the noise model are calculated. At the sixth step, a ratio of the difference of the probability of the word to be recognized to the probability of the word is calculated using statistics, and a mean and a variation are calculated. |
