| 摘要 |
PROBLEM TO BE SOLVED: To recognize voice with high precision with respect to a plurality of users. SOLUTION: A model adapting part 12 detects an optimum conversion function for adapting an input voice to an acoustic model among one or more conversion functions based on conversion results which are obtained by converting the input voice through the use of one or more conversion functions stored in a conversion matrix storage part 13 and assigns input voice to the optimum conversion function. Moreover, the model adapting part 12 updates the conversion function to which the new input voice is assigned through the use of the whole input voices assigned to the conversion function. In the meantime, a selecting part 14 selects the conversion function to be used for converting the input voice among one or more conversion functions stored in the conversion matrix storage part 13. A converting part 5 converts the input voice by the selected conversion function. Then a matching part 6 matches the input voice converted by the conversion function with an acoustic model. |
