To provide a speech recognition word addition method which is capable of appropriately adding an unregistered word likely to be spoken, at a low calculation cost in the case that known words used in simular context do not necessarily exist.
A cooccurrence frequency vector in a relevant document of an unregistered word o and a word frequency vector being an uttered word appearance frequency included in a speech recognition result are obtained, and similarity between both the vectors is obtained as a context similarity r(o), and a reciprocal of the number of registered words in a target class after registration of the unregistered word o is obtained as an intra-class average language probability pCo, and an intra-class appearance probability p(o|Co) of the unregistered word is obtained on the basis of the intra-class average language probability pCo, a reference value r(ob) of context similarity, and the context similarity r(o), and the unregistered word o and the intra-class appearance probability p(o|Co) are outputted.
YAMAGUCHI YOSHIKAZU
JP2009271465A | 2009-11-19 | |||
JP2013171222A | 2013-09-02 | |||
JP2013109125A | 2013-06-06 | |||
JP2010282199A | 2010-12-16 | |||
JP2009271465A | 2009-11-19 | |||
JP2013171222A | 2013-09-02 |
US20100332231A1 | 2010-12-30 | |||
US20070005360A1 | 2007-01-04 |
JPN6014032119; 山畠 祥子 Shoko YAMAHATA: '関連ドキュメントとの文脈類似度と音声認識信頼度に基づく語彙自動獲得' 日本音響学会 2012年 春季研究発表会講演論文集CD-ROM [CD-ROM] , 20120315, p.243-244, 社団法人日本音響学会
CSNJ201210012557; 山畠 祥子 Shoko YAMAHATA: '関連ドキュメントとの文脈類似度と音声認識信頼度に基づく語彙自動獲得' 日本音響学会 2012年 春季研究発表会講演論文集CD-ROM [CD-ROM] , 20120315, p.243-244, 社団法人日本音響学会
Yukio Nakamura
Yoshimura Munehiro