The Research Of Speech Recognition Based On Covering Neural Network Ensemble

Language is one of the most important means of exchanging information among the mankind. With the continuous developing and application of electronic computer and the increasing improvement of artificial intelligent, people expect to make the computer recognize human language. This requirement makes the technique of speech recognition have immense space to develop. The technique of speech recognition is becoming the key technique of man-machine interface in information technique. It enables people to get rid of the keyboard and operate by speech. However, speech recognition is related to a lot of factors, which must be considered simultaneously. As a cross-discipline, it is related to a lot of subjects such as signal processing, pattern recognition, artificial intelligent, computer science, language, recognition, etc. Therefore the development status in speech recognition is still far from our final aim and some associated technological difficulties still need to be conquered.This paper focuses on thorough analyses and investigations of the main process of speech recognition. First of all, the speech signals inputted must be pre-processed to ensure that the system will obtain an ideal object to process. In the stage of drawing the character, the usually used character parameters LPCC and MFCC and so on, are described in this paper. In the recognition stage, some techniques are introduced such as VQ, DTW, HMM and so on, and ANN is the emphasis of introduction.ANN is an information process that is used for simulating biologic neural systems. It is widely used in association-memory, nonlinear-mapping, classify and recognition , optimize computing , knowledge processing, etc. ANN shows great potential in speech recognition, large-scale mode recognition. But the generalization ability of single network is not perfect in dealing with this large-scale mode recognition.ANN Sphere neighborhood model is the model of neural network by which the training problem of neural network may be transformed into the geometrical covering problem of a point set based on n+1 dimensional sphere. It has obvious geometrical representation. Neural network ensemble can significantly improve the generalization ability of learning systems through training a finite number of neural networks and...