Research And Design Of Embedded Speech Recognition System

Language is an important method of human communication with the outsideworld,being able to communicate with the machine naturally and with nobarriers,meanwhile,let it react accordingly,has been the human pursuit of dreams for along time.With the development of computer technology and wide application ofartificial intelligence machines,more and more intelligent products began to go intopeople’s lives.Due to the disadvantage of large volume and high prices of traditionalPC,it’s difficult for them to apply into the pratical products.Combining the speechrecognition system with cheap embedded system of good portability、low powerconsumption,become the ideal choice for people no doubt.This paper puts forward the optimization and improvement schemes based on afurther study of speech recognition algorithm,and make use of MATLAB softwareplatform to validate and make a simulation,finally combining the speech recognitiontechnology and embedded technology,have designed a small embedded speechrecognition system based on SPCE061A.The main research work of the paper is asfollows.Firstly,this paper has carried on the detailed introduction and analysis about thekey technologies of speech recognition,including the research methods of short-timeenergy and zero-crossing rate characteristics in the scope of speech signal timedomain、extraction algorithm of allinear prediction cepstrum coefficient and Melfrequency cepstrum coefficient.It studies the double threshold endpoint detectionalgorithm emphatically,has put forward energy zero difference detection algorithmthrough calculating the difference of short-time energy and zero-crossing rate,andmakes the simulation and comparation of this two algorithms,proves that the lattermakes a certain improvement on the accuracy and efficiency of the algorithmdetection.In addition,has completed the speech signal characteristic parametersextraction with Mel frequency cepstrum coefficient and first order differencecoefficient.Secondly,in terms of speech recognition algorithm,analyzes and compares theadvantages and disadvantages of the Dynamic Time Warping,Hidden Markov Model and Artificial Neural Network algorithm,mainly discusses the recognition algorithmbased on Hidden Markov Model and studies its implementation principle andlimitations,optimizes the actual application method,uses compound K-meansclustering algorithm to make a parameter revaluation,make logarithmic processing ofViterbi recognition algorithm.Experiments prove that the operation efficiency andrecognition rate of the algorithm is improved,it reduces the demand for systemconfiguration and has better portability.Finally,apply the improved recognition algorithm to the hardware platform basedon SPCE061A single chip computer,respectively completes the design of thehardware circuit and software program,implements a valuable embedded speechrecognition system.The system test shows that it has good feasibility andpracticability,according to the specific requirements of practical application,it can beported to vehicle terminal system、smart home system and entrance guard system,etc.