Research On Speech Recognition System For Equipment Controlling

In modern times, machines are in everywhere of people's life. Now the keyboard is the primary HMIs(Human and Machine Interfaces). The keyboard control mode had too many inconvenience, such as the user must use his eyes and hands at the same time, and the keyboard control is difficult to learn to some people. Speech communication is the most simple and convenience way for people to know each other. If we can control the machine by voice it will make the great convenience. In order to implement speech communication between people and machine, we must improve the utility of speech recognition. The paper is working on the research of speech control technology.This paper analysed and designed a speaker-dependent, isolated word and small vocabulary speech control system. This paper analyses and compares the advantage and the disadvantage between the Dynamic Time Wrapping (DTW) and the Hidden Markov Model (HMM).Because we are to design a speech control system which is for special operator and the number of command is less than 50. Dynamic Time Wrapping can meet the require very well, So we choose the Dynamic Time Wrapping to be the algorithm for speech recognition. The idea of the algorithm is to use the dynamic programming to align and normalize the sequences of acoustic features. The process of speech recognition goes along as following: pre-emphasis,frame blocking,windowing, endpoint detection,MFCC feature extraction, temporal cepstral derivative.This paper designed a speaker-dependent, isolated word and small vocabulary speech recognition system which is based on DTW algorithm. And researched the key techniques and programming the software system. This paper stress on endpoint detection which is very important to isolated-word speech recognition system. This paper also make a point research of several feature parameters, such as linear prediction cepstrum coefficient (LPCC),Mel-Frequency Cepstrum Coefficient(MFCC). The Mel- frequency cepstrum coefficient is adopted to be the speech characteristic parameter. Aimed at two obvious defects of DTW ,one is sensitivity to the endpoint ,second is the tremendous operation, this paper make a modification of the tremendous operation, this paper make a modification of the traditional DTW model. It made the model more speedy and make good effect. This paper used two threads to design the speech recognition software.At last this paper simulate the speech recognition system in matlab. The speech data used in model are all from the speech signal of the microphone and computer audio card .This paper stressed on the simulation of the number and on, off etc. and got a approving result.