| 摘要 |
PURPOSE: To improve the recognition rate by rejecting distinct errors. CONSTITUTION: The cumulative scores of the results of word recognition are extracted in a step S1. A first candidate and a second candidate are selected in a step S2 from the min. scores among the extracted scores. Next, the respective cumulative scores of the first candidate and the second candidate are obtd. in a step S3. The frame number of the utterance data from the step S2 is obtd. in a step S4. The difference in the cumulative scores between the first candidate and second candidate obtd. in the step S3 is adopted in a step S5 and whether the difference is '0' or not is judged. The distinct error, etc., among the erroneously recognized scores are rejected in a step S6 in the case of YES. The processing of a step S7 is executed in the case of NO in the step S5. The step S7 rejects the distinct error, etc., in the same manner as in the step S6 if the difference in the cumulative scores between the first candidate and the second candidate is >='1' to <='3' and if the first cumulative score is <=1/2 the frame number obtd. in the step S4. |
