| 摘要 |
<p><P>PROBLEM TO BE SOLVED: To solve the problem that the accuracy of speech recognition employing an acoustic model having insufficient learning data has been low with respect to a minor language. <P>SOLUTION: In an acoustic model learning device, a first acoustic model in a first language, a second acoustic model in a second language, a first pronunciation dictionary in the first language, and second phoneme-relevant information in the second language are selected, a phoneme identifier included in a phoneme identifier string which first word pronunciation information has is substituted with a selected phoneme identifier which second phoneme-relevant information has, to constitute temporary first word pronunciation information, one or more recognition results are acquired by performing speech recognition processing using the first acoustic model and a temporary first pronunciation dictionary obtained by adding the temporary first word pronunciation information to the first pronunciation dictionary, substitution count information of the temporary first word pronunciation information is acquired, and new first phoneme-relevant information which associates phoneme information in the second language and substituted phoneme identifiers of the first word pronunciation information with each other is constituted in the case of substitution count information being large and is stored in the first acoustic model. By the acoustic model learning device, acoustic models in many languages can be efficiently constructed. <P>COPYRIGHT: (C)2012,JPO&INPIT</p> |
