LANGUAGE DISCRIMINATION DEVICE

摘要

PURPOSE:To improve the language discrimination capability by grasping the differences in phoneme pronunciations.spectrum structures in languages through the use of an ergodic HMM for each language and also grasping the phoneme array structure of the language by the ergodic HMM condition group which is common to all languages. CONSTITUTION:A feature extracting section 1 converts text audio signals into the series of feature vectors. An all language common ergodic HMM generating section 5 generates an all language common ergodic HMM from the feature vector series for all language extracted by the section 1 using the text sounds of various kinds of languages as learning sounds and the HMM is stored in a storage section 6. An optimum condition series calculating section 7 obtains an optimum condition series of the all language common ergodic HMM corresponding to the feature vector series. A trigram generating section 10 generates the trigram of an optimum condition series and a trigram model corresponding to the language generated by using the text sounds generated by trigram storage sections 11 to 13 is stored.