| 摘要 |
PROBLEM TO BE SOLVED: To prevent an erroneously recognized word from being recurred by estimating the erroneously recognized word being in a recognized result. SOLUTION: A recognized result with respect to the most similar word or a string of single words which is to be outputted from the matching part of a voice recognition system is inputted to a recognized result input means 1. Whether the recognized result inputted to the input means 1 is a correct interpretation or not is confirmed in a confirming means 2. When the recognized result is confirmed to be an incorrect interpretation in the means 2, the matching score among respective words constituting the recognized result and the partial string of an input frame in which the word is considered to exist is extracted in a matching score extracting means 3. After the score (for example, ith word of words) extracted in the extracting means 3 is subjected to a comparision in a score comparator means 4, when the ith word is judged that the recognition of the word is the worst and the word is eliminated from a dictionary for recognition in an eliminating means 5. |
