| 摘要 |
<P>PROBLEM TO BE SOLVED: To provide a voice recognition device capable of outputting a character string expressing a voice represented by voice information with high accuracy while preventing an amount of information held in a storage device from becoming excessive. Ž<P>SOLUTION: This device 1 obtains a per-frame voice obtained by dividing voice represented by received voice information into each time frame sequentially from the start end of the voice to a terminal end one by one. The device obtains a candidate for an element string representing the per-frame voice which is obtained based on the obtained frame by a frame voice and a word internal assumption which is an assumption of an element string constituting a word representing a voice and stored in a storage device. The device connects the obtained candidate to the word internal assumption. When the word internal assumption represents a word, the device connects the word to a word terminal end assumption which is an assumption of a word string representing a voice and stored in the storage device. Upon receiving an erase designation, the device erases a part of the word terminal end assumption which can generate a word string as a character string representing a voice represented by the stored and received voice information. Ž<P>COPYRIGHT: (C)2010,JPO&INPIT Ž
|
