PROBLEM TO BE SOLVED: To provide a text creation device for acoustic model learning, that is capable of extracting a text containing "a phoneme that ought to be learnt".;SOLUTION: A voice recognition processing section recognizes a voice in developing voice data inputted from outside by referring to a language model and an existing acoustic model, and outputs a recognition result text and phonemic series information. A recognition result tabulation section calculates a phoneme recognition rate from the phonemic series information and a correct interpretation text of the developing voice data. A weak-point phoneme extracting section extracts, as a weak-point phoneme, a phoneme of which the phoneme recognition rate is equal to a threshold level or below, and produces a weak-point phoneme list. A supplementary candidate text corpus stores a large amount of text being a candidate text for acoustic model learning. A weak-point phoneme containing text selecting section refers to the weak-point phoneme list to select a text containing the weak-point phoneme from the supplementary candidate text corpus, and outputs the selected text as an acoustic model learning text.;COPYRIGHT: (C)2014,JPO&INPIT
展开▼