摘要 |
PROBLEM TO BE SOLVED: To improve a voice recognition rate even though a double consonant character and a contracted sound character are erroneously registered into a dictionary database. SOLUTION: Correspondence between a phoneme and a phoneme pattern is beforehand registered in a phoneme pattern database 34. In a voice recognition engine 35, (1) plural character string candidates which have voice patterns having a high degree of similarity to input voice are obtained while referring to a voice dictionary database 33, (2) when the double consonant character or the contracted sound character is included in each of the candidate character strings, the size of the double consonant character or the contracted sound character is changed to generate a virtual candidate character string, (3) the virtural candidate character string is decomposed into phonemes and the phoneme pattern corresponding to each phoneme is obtained while referring to the phoneme pattern database and the voice pattern of the virtual candidate character string is generated by synthesizing the obtained phoneme pattern and (4) character string having a highest degree of similarity to the input voice is selected among the candidate character string and the virtual candidate character string and the input voice is recognized. |