Title:
【発明の名称】音声理解方式
Document Type and Number:
Japanese Patent JP3024187
Kind Code:
B2
Abstract:
PURPOSE:To modify a network which represents grammar by identifying the step of a plan model by voicing and storing the context of a talk, setting the expectancy of proposition contents regarding the realization of respective partial purposes, and using this expectancy. CONSTITUTION:The recognition result outputted by a pattern recognition processing part 407 which is an HMM voice recognition module is held in a voicing proposition contents buffer 409 and a talk processing control part 406 tries to make a plan skimmer instant and detailed. The voicing type of a proposition expression held in a buffer 409, on the other hand, is held in a voicing type storage part 401. The control part 406 finds a set of possible voicing types for next voicing and a set of arcs corresponding to them from the voicing type of current voicing contents held in the storage part 401; and the expectancy of proposition contents corresponding to a proposition pattern which is made concrete is set low and the expectancy of proposition contents corresponding to a proposition pattern which is not concrete is sets high, thereby weighting the set of arcs with the expectancies. The grammar network which is weighted is used by a processing part 407 to recognize next voicing.
Inventors:
Shinsuke Sakai
Application Number:
JP24363390A
Publication Date:
March 21, 2000
Filing Date:
September 13, 1990
Export Citation:
Assignee:
NEC
International Classes:
G10L15/18; G10L15/12; G10L15/193; G10L15/22; (IPC1-7): G10L15/22; G10L15/18
Other References:
・山岡ら「文脈を考慮した音声認識結果絞り込み手法」情報処理学会研究報告 ＮＬ７８−１６，ｐｐ１２１−１２８（１９９０）
・野村ら「タスクを限定した対話文音声理解システム」信学技報ＳＰ８８−２５，ｐｐ３１〜３７（１９８８）
・野村ら「タスクを限定した対話文音声理解システム」信学技報ＳＰ８８−２５，ｐｐ３１〜３７（１９８８）
Attorney, Agent or Firm:
Naoki Kyomoto (2 outside)