Title:
SPEECH RECOGNIZING METHOD AND RECORDING MEDIUM WHERE SPEECH RECOGNIZING PROGRAM IS STORED
Document Type and Number:
Japanese Patent JP3454711
Kind Code:
B2
Abstract:
PROBLEM TO BE SOLVED: To provide a speech recognizing method which enables a fast processing by making the shift width of a speech analysis window different for the learning of a model and for the recognition of an input voice.
SOLUTION: A speed coefficient is inputted from a speech coefficient input step 77. Analysis conditions for HMM(hidden Markov model) generation are read out of an HMM memory 78. An analysis window first width among the analysis conditions is multiplied by the speed coefficient to determine an analysis window shift width for recognition in an analysis window shift width determination step 72. A voice signal to be analyzed inputted from a voice input step 71 is analyzed in a feature quantity calculation step 73 under the determined conditions. Dynamic feature quantities among feature quantities are corrected by a dynamic feature quantity correction step 74 so as to meet the analysis conditions for HMM generation. Likelihood calculation is performed in a likelihood calculation step 75 by using the feature quantities and HMM, and the HMM category having the largest likelihood is outputted as a recognition result from a recognition result determination step 76.
Inventors:
Satoshi Takahashi
Application Number:
JP11573698A
Publication Date:
October 06, 2003
Filing Date:
April 24, 1998
Export Citation:
Assignee:
Nippon Telegraph and Telephone Corporation
International Classes:
G10L15/10; G10L15/02; G10L15/14; (IPC1-7): G10L15/02; G10L15/10; G10L15/14
Domestic Patent References:
| JP1106096A | ||||
| JP6091400A |
Attorney, Agent or Firm:
Masatake Shiga
Previous Patent: FLOW MEASURING DEVICE
Next Patent: COLOR FILTER AND ITS MANUFACTURE, AND LIQUID CRYSTAL DISPLAY DEVICE
Next Patent: COLOR FILTER AND ITS MANUFACTURE, AND LIQUID CRYSTAL DISPLAY DEVICE