Research Of Speech Recognition And Its Application In The Speech Error Identifying System

Since the 1950s,the technology of speech recognition (SR) has been developing rapidly. As an interdisciplinary field, SR has developed as an integrated theory.On the other hand, the SR systems running on PC have developed so well and have made some success.Although lots of SR's products come forth in succession,but in many special fields,because of the particularities of the circumstance and industry,we often need to exploit the system specially.During their study of English,students often lack standard and effective training of spoken English.To solve the problem,depending on some research and improvement of the SR technology, the Speech Error Identifying System is exploited.The system could teach the students how to pronounce,grade their pronouncation and even identify the errors.Firstly, according to the basic models of SR system,this thesis analyzes the fundamentions of every part(including pretreatment,characteristic parameters distilling,template production and template matching),such as the speech data signal processing and the mode recognition.Secondly, this thesis mainly deals with the theory of search algorithn for continuous SR(CSR) and optimizes it.In this thesis,we discuss the principles and implementation of beam pruning search algorithm in depth,and discuss the state level pruning and word level pruning.The limits of beam pruning search algorithm are also analyzed in the thesis. In this thesis,we improve pruning search search algorithm from tow aspects.On the one hand,we present a new adaptive beam pruning search algorithnl on the state level referred to adaptive beam pruning search algorithm based on variation of active model numbers. On the other hand,we set the adaptive beam pruning search algorithnl on the word level depending on the max probability score and the average probability score on currently search path.Lastly, with the combination of the adaptive beam pruning search algorithm on the two levels and SR engine, this project is exploited using the SAPI interfaces with the optimized SR engine.The system has three functions:speech synththesis,speech recognition and speech error identifying.The test of the system shows that it has good ability of CSR.After imitating and training to the students,the identifying rate could reach 81.2%.This also means a success of the combined adaptive beam pruning search algorithm on the state level and the wod level.