| 摘要 |
PURPOSE:To save the processing quantity of the case in which vocabulary does not increase by executing word matching processing just once even in spite of the presence of plural state transitions in which words appear. CONSTITUTION:The inter-vector distance for the time (i) of respective input patterns and the cumulative matching distance at the time i-1 just before the same are subjected to the matching with the respective words by prescribed equations in a word matching processing section 5. The cumulative matching distance at the time (i) is stored into a buffer 6. The initial setting of the word matching is executed in a word boundary processing section 8 and the cumulative distance for the state transition which is not subjected to the matching is calculated from the result of the matching and is stored into a word boundary buffer 10. The word array of the result of recognition is obtd. with the beginning point and state transition of the words as back frame by reading out the word boundary buffer 10 in a word array forming section 11 when the repeating processes to the respective time (i) of the inputs reach the ending time. The increase in the processing capacity is prevented in this way in spite of complicated compositions unless the vocabulary increases. |
