| 摘要 |
<p><P>PROBLEM TO BE SOLVED: To provide a voice recognition apparatus for reducing correction load in a voice recognition error. <P>SOLUTION: The voice recognition apparatus is provided with: a semantic relation storage part 124 for correspondingly storing semantic relation between words and relevancy; a key input receiving part 111 for receiving a voice input; a candidate generation part 114 for generating recognition candidates and likelihood; a first candidate selection part 115a for selecting a recognition result from first recognition candidates corresponding to the first voice; a second candidate selection part 115b for extracting word recognition candidates corresponding to a word to be corrected and word recognition candidates of a key word from second recognition candidates corresponding to the second voice, which include the word to be corrected in the first voice and the key word for correction, acquiring relevancy corresponding to semantic relation between the extracted recognition candidates and selecting a recognition result on the basis of the relevancy; a correcting position specification part 116 for collating the recognition results of the first and second voices and specifying the correcting position of the first voice; and a correction part 117 for correcting the correcting position. <P>COPYRIGHT: (C)2008,JPO&INPIT</p> |
