| 摘要 |
<p>PROBLEM TO BE SOLVED: To be able to use an expression system of '○×Δof○', which is an often used means in the communication between human beings when KANA (Japanese syllabary) characters are to be transmitted accurately, for voice recognition. SOLUTION: When a conventional large vocabulary voice recognition system is used as it is, a series of simple sentences such as '○×Δof○' is regarded as one word in lieu of individual words '○×Δ', and a sentence having continuously this word 'of○' is processed, the size of the network is greatly increased. Therefore, an original already registered word portion and the portion after 'of' are separated as units for making a network and then, processes are conducted. In this case, the network of the portion after 'of' has enough a small size to only express all Japanese syllabary characters, the number of registered vocabulary has no relationship and the network of this portion occupies a very small portion of the entire grammar.</p> |
