| 摘要 |
PURPOSE:To improve the recognition rate by utilizing the extent of politeness in a history of user's speech up to now. CONSTITUTION:A politeness arithmetic part 103 calculates the extent of politeness from a 1st score for each recognition result candidate sequence based upon a standard pattern for speech recognition stored previously in a standard pattern storage part 107 as the output of a speech recognition part 102 and the 1st score, and grants the calculation result to each recognition result sequence, and outputs the result to a recognition candidate selection part 104. A speech history storage part 106 is stored with a history of a 2nd score showing the extent of politeness of the recognition result of each sentence speech that the user has inputted to the system so far. A recognition candidate selection part 104 selects a recognition result that is considered to be an optimum result among the recognition result candidate sequences outputted by the speech recognition part 102 in consideration of recognition scores of respective candidate sequences, the 1st score showing the extent of politeness, and the 2nd score showing the extent of politeness of speech obtained so far in the speech history storage part 106, and then outputs the selected recognition result to a recognition result output part 105. The recognition result output part 105 stores the scores of the recognition results as a 2nd score in the speech history storage part 106. |
