Study On Speech Recognition Key Technologies And The Realization Of The System

Speech recognition is to make machines understand human's speech, and make the correct response. The ultimate purpose of speech recognition is to achieve a natural communication between human and machines. With the development of science, speech recognition technology is gradually forming a more complete theoretical system, and more and more related products appear, but because of different requirements in variuos fields, it needs to focused development. The two important directions of speech recognition technology are large vocabulary recognition based on PC and embedded speech recognition, and there are broad market prospects. On the basis of studying the key technologies of speech recognition, I put forward a new recognition algorithm——GA_DTW. Compared with the traditional DTW algorithm, the new algorithm has more excellent global search ability and parallel computing characteristics. The results of experiment show that the effectiveness of the algorithm, and the recognition rate of isolated words is 95.07%. Based on the GA_DTW algorithm, a small embedded speech recognition system is designed, and use the algorithm on the system, and obtain a satisfictory result through the experiments. This paper's primary work as following:(1) Analyze the basic theory of speech recognition, including the basic composition of the voice, properties and digital model of Chinese speech, voice signal sampling, pre-emphasis, windowing, framing, endpoint detection and parameters extraction. For the most important speech endpoint detection, an improved endpoint detection algorithm called Dual Dynamic speech endpoint detection algorithm is brought forth, and the experiments show that the algorithm has better detection performance. After analyzing the features of Linear Prediction Cepstrum Coefficient(LPCC) and Mel-Frequency Cepstrum Coefficient(MFCC), I choose MFCC as the characteristic parameter of this study, and extract the parameters.(2) Study the current three mainstream speech recognition algorithms: DTW, HMM and ANN, analyze their principles, characteristics and realization processes, and make some speech recognition experiments with the DTW. By comparing with and analyzing the characteristics of this three algorithms, according to the actual situation of this paper, I choose DTW as the focus of the study , and bring forth using of Genetic Algorithms to improve it.(3) On the basis of analyzeing Genetic Algorithm, use its excellent global search ability and the characteristic of parallel computing to improve the traditional DTW algorithm, and bring forth a new speech recognition algorithm called GA_DTW. In the design of GA_DTW, it focuses on studying the realization mechanism, encoding, fitness function design, population initialization, selection mechanism, crossover operator, mutation operator and termination celue. And do some experiments, the results prove the new algorithm's effectiveness and efficiency.(4) Based on the GA_DTW algorithm, design a small embedded speech recognition system at the core to the microcontroller of SPCE061A. The details of this design are analyzed from the aspects of hardware and software, and use C language to realize GA_DTW algorithm. The tests of this system get a satisfactory result.