Research On Hmm-based Speech Recognition System Of The Robot

With continued research on intelligent robots, a higher demand of the human-computer interaction is needed. .As speech recognition technology and robot control technology combined, the human can control the robot through voice signals, so the intelligent human-computer interaction becomes a focus of research on intelligent robot.In this speech recognition robot control system, we require voice commands which fit some specific environments to control the behaviors of the robot. In the beginning of the study, the control commands are some words, so the position of this paper is to design a non-specific human isolated word speech recognition system on robot.This paper deeply analysis of the basic principles and basic processes of the speech recognition system, and deeply studies the endpoint detection algorithm based on energy and zero-crossing rate and feature extraction algorithm based on linear prediction(LP) and Mel frequency spectrum. The paper also deeply studies the three classical algorithms of the hidden Markov model (HMM): forward-backward algorithm, Viterbi algorithm, Baum-Welch algorithm, and theory and process of the isolated word speech recognition system based on HMM.On given robot platform in laboratory, we used the way of modular software design to design software. The whole speech recognition and control system on robot can be divided into three module, the human-machine interface module and the speech recognition module were designed and implemented in details in this paper. The human-machine interface module was achieved by Qt graphics library, the acoustic model of the speech recognition system was created by HTK toolkit which developed by the Cambridge, the recognizer of the speech recognition used the speech recognition engine Julius which developed by Kyoto University and IPA in Japan.