Title:
【発明の名称】つづり-発音記号変換装置
Document Type and Number:
Japanese Patent JP3004753
Kind Code:
B2
Abstract:
PURPOSE:To efficiently make a neural network execute learning without requiring any specialists or the like. CONSTITUTION:This device is composed of an input character window 30 to input the spelling characters of a word and the phonetic symbols of the spelling characters and to segment/output N pieces of partial character strings, N pieces of character delays 31 to output spelling delayed for each character while receiving the spelling characters from the input character window 30, N pieces of neural network parts 40 composed of element neural networks 41 to input outputs corresponding to these character delays 31 and learning parts 42 to execute learning operations to the element neural networks 41, teaching signal generation part 32 to generate a teaching signal while receiving the phonetic symbol from the input character window 30 and to apply the teaching signal to the learning part 42 of the element neural network 41, weight coefficient part 33 to execute prescribed weighting to the respective outputs while receiving the outputs of the respective element networks 41, and final output part 34 to output a final phonetic symbol sequence (a) while receiving the output of the weight coefficient part 33.
Inventors:
Tatsuro Matsumoto
Yukiko Yamaguchi
Yukiko Yamaguchi
Application Number:
JP5698391A
Publication Date:
January 31, 2000
Filing Date:
March 20, 1991
Export Citation:
Assignee:
富士通株式会社
International Classes:
G06F15/18; G06F17/28; G06G7/60; G06N3/00; G06N99/00; (IPC1-7): G06F17/28; G06F15/18
Domestic Patent References:
| JP4109299A | ||||
| JP227395A | ||||
| JP1238697A |
Attorney, Agent or Firm:
Fujishima Ijima (1 outside)