Study of Speech Command Recognition

• Main Author: 江智堯
• Other Authors: 王朝興
• Format: Others
• Language: zh-TW
• Published: 2009
• Online Access: http://ndltd.ncl.edu.tw/handle/66990029281793057214

碩士 === 國立彰化師範大學 === 電機工程學系 === 97 === In this paper, Mel-frequency cepstral coefficients(MFCC) is as the feature parameter of speech., and Hidden Markov Models as the acoustics model. Generally, speech recognition system is divided into language and acoustics parts. This paper is intended to investigate Speech Command Recognition. This paper adopts three methods. First method uses the model structure of the whole word into long, middle and short state to look for the best model. The second method revises the whole word model to the initial and final model. It aims to look for the best state number and to modify the mixture number that is promoted for recognition. The third method uses the phoneme to acoustics unit to look for the best state and to modify the mixture number that is promoted for recognition. The third method uses the phoneme to acoustics unit to look for the best state and modify mixture number that is promoted recognizes. There are experimental results that can know the recognize rate of second of method comparatively high. we have contructed a real-time voice command recognition system that can be used to control a robot.