SOUND MODEL STRUCTURING DEVICE, SPEECH RECOGNITION DEVICE, SOUND MODEL STRUCTURING METHOD, AND PROGRAM

摘要

PROBLEM TO BE SOLVED: To solve such a problem that precision of speech recognition processing using sound models whose learning data are not sufficient is low.;SOLUTION: Sound models of many languages can be efficiently structured by a sound model structuring device including a sound model storage part which stores sound models of one language, a speech data storage part which stores speech data of one or more phonemes of one or more other languages, a phoneme frequency acquisition part which obtains a phoneme frequency of each phoneme identifier, a determination part which applies each phoneme frequency to conditions associated with the phoneme frequency to determine whether the phoneme frequency is lower as conditions are met more, a speech data acquisition part which obtains one or more speech data having phoneme identifiers corresponding to one or more phoneme identifiers such that phoneme frequencies are lower as conditions are met more, a learning part which obtains one or more feature quantities from the obtained one or more speech data to obtain one or more pieces of other-language phoneme information having the one or more feature quantities and phoneme identifiers, and a sound model structuring part which stores the one or more pieces of other-language phoneme information.;COPYRIGHT: (C)2012,JPO&INPIT