SPEECH RECOGNITION METHOD, DEVICE THEREFOR AND SPEECH OPERATED CONTROLLER

摘要

PROBLEM TO BE SOLVED: To provide a speech recognition method with which a high performance speaker adaptation can be carried out with a few utterance of the speaker. SOLUTION: By using a user's speech which is inputted from a speech input part 201 and acoustic analysis is carried out in an acoustic analysis part 202, a pattern most suitable for the user's speech is selected in a pattern selection constitution part 204 classified by the feature, from the pattern classified by the feature stored in a pattern storage part 203 which preliminarily classifies the learning speakers according to the feature and holds two or more learned learning patterns, and a succeeding speech recognition performance can be improved with a small burden to the user by determining the spectrum frequency distortion coefficient for correcting the difference in the frequency of the spectrum caused by the vocal tract length of the speech of the pattern learning speaker and the input speaker with a speaker adaptation processing part 205.