发明名称 WFST CREATION DEVICE, SPEECH RECOGNITION DEVICE, SPEECH TRANSLATION DEVICE, WFST CREATION METHOD, AND PROGRAM
摘要 PROBLEM TO BE SOLVED: To solve the problem that addition of a new word to a WFST requires reconstruction of a WFST.SOLUTION: A new word can be easily added to a WFST by a WFST creation device 1 including: a CLWFST storage unit 11 in which a WFST of CL being a WFST obtained by combining a phoneme environment-dependent WFST(C) being a WFST for converting triphone phoneme strings to single phoneme strings and a word dictionary WFST(L) being a WFST for converting single phoneme strings to word strings can be stored; a GWFST storage unit 12 in which a WFST of G being a WFST(G) of language models having words and weights can be stored; an accepting unit 13 which accepts a single phoneme string of a word and a character string of the word; and a new word storing unit 14 which constitutes a state transition with the single phoneme string of the word as input and with the character string of the word as output and stores it into the GWFST storage unit 12.
申请公布号 JP2014106272(A) 申请公布日期 2014.06.09
申请号 JP20120257341 申请日期 2012.11.26
申请人 NATIONAL INSTITUTE OF INFORMATION & COMMUNICATIONTECHNOLOGY 发明人 DIXON PAUL RICHARD;HORI CHIORI;KASHIOKA HIDENORI
分类号 G10L15/193 主分类号 G10L15/193
代理机构 代理人
主权项
地址