Title:
SPEECH RECOGNIZING SYSTEM
Document Type and Number:
Japanese Patent JP3460723
Kind Code:
B2
Abstract:
PURPOSE: To provide a speech recognizing system hard to receive the influence of a beginning of word with unstable utterance and that due to the erroneous segmentation of an input audio interval and capable of reducing the number of candidates set as the target of recognition.
CONSTITUTION: When a voice to be set as input is uttered, a unit string evaluation means 4 performs the matching of the candidate of each unit string supplied from a unit string generating means 3 with an input voice from a recognition start point decided in a recognition decision means 1, and stores an evaluation result in a unit string storage means 2. The unit string generating means 3 connects a new unit to the unit string stored in the unit string storage means 3, and supplies it to the unit evaluation means 4. Such processing is repeated until no new unit is connected, and the unit string with the highest evaluation value out of the unit strings stored in the unit string storage means 2 at that time is set as the final recognition result.
Inventors:
Hitoshi Iwamida
Application Number:
JP12627992A
Publication Date:
October 27, 2003
Filing Date:
May 19, 1992
Export Citation:
Assignee:
富士通株式会社
International Classes:
G10L15/10; G10L15/18; G10L15/183; G10L15/193; G10L15/22; (IPC1-7): G10L15/10; G10L15/18
Domestic Patent References:
| JP3141398A | ||||
| JP6026400A | ||||
| JP5978400A | ||||
| JP6027000A | ||||
| JP61261798A | ||||
| JP3214198A | ||||
| JP457099A | ||||
| JP5108090A |
Other References:
【文献】古井,ディジタル音声処理,東海大学出版会,日本,1985年 9月,p.12
Attorney, Agent or Firm:
Shunichiro Nagasawa