Research Of Isolated Word Speech Recognition System And Its Application Based On HMM

There are three means for people in their daily lives to achieve mutual communication,suchas use gestures, expressions and speech.And speech is one of the most frequently used means.This is because the voice with high efficiency and convenience features. the characteristics ofthe speech recognition technology will become the main means of communication between manand machine. The speech recognition technology is mainly to extract speech feature informationand set a model, and then to the pronunciation model training to get training module, test toextract speech characteristics, at last use the features of the training template you get and thestandby to test of the speech features to match.As the development of speech recognition technology, more and more technicists set thefocus on the area of speech control. The key of speech control is that the robot does not only needto get the meaning of the orders, but also can know what to do according to the orders. Based onthe speech control technology, small-scale vocabulary speech recognition system is discussed inthis thesis.Research shows that,in a very short time interval speech signalcan be seen as a stable.So thespeech signal can be described by the linear model in a very short time,But on the whole it seemsthe speech signal is changing with time.This requires the representation of the signal model,themodel parameters must also be instant change. Therefore,people think of such a method:thespeech signal is divided into a number of short time interval,and then in the short time series withthe linear model.Finally, according to the time order of the short time series with the linearmodel,which constitutes a Markov chain.Markov chain describes the statistical relationshipcorresponding to the state and observation value.The basic idea of HMM is modeling,the model is the transition probability andtheobservation probability based on speech rate and output acoustic changeprimitives usingstate.Two main aspects are included in this thesis. On the one hand, the principle of speechrecognition and the speech acoustical model (Hidden Markov Model) are discussed. On the otherhand, an introduction of control dolly platform and software platform are discussed in the thesis.The programming of the software platform is based on VC, good performance is achieved inexperiments.