The Noise Speaker Dependent Speech Recognition System Base On PCNN

AS the computer has become more and more portable, the environment of that has been more complicated, human desire to communicate with computer in a portable, natural, effective way without the constraints of keyboard. In this situation, speech recognition technology has developed, which controls the computer to implement the order according to the speech. It can bring the human much convenience. At present, it has been applied extensively in human life, such as household electronics, intellectual toys, the database voice inquiry of the business system, industrial voice control, autodialing of telephone and telecom system, etc. Because of the advantages of speech recognition technology, it is of much possibility to develop into the next interface of the operation system.Although speech recognition technology has made great progress, there are still many problems needed to be solved. One of them is the speaker dependent speech recognition system in the noisy environment. The system has high recognition rate in laboratory environment, but people inevitably suffer from the noise distraction in the speech communication process. The distraction can reduce the system recognition rate significantly. In order to improve the speaker dependent speech recognition rate in the noisy environment, this paper uses spectral subtraction to denoise, for it is easier to be implemented and calculated.At the same time, for effectively suppressing the'musical noise'which is produced during the spectral subtraction, this paper makes some improvement in the spectral subtraction. The experimental results show that the improved spectral subtraction effectively denoises and enhances the recognition rate greatly, the system doesn't produce excess'musical noise'as a result of using the improved spectral subtraction.Feature parameters are the fundament of the speech recognition system,they must express the information of the speech signal fully and accurately. How to choose the effective and reliable feature parameters is one of the key points of the system, it directly influences the inferiority and recognition rate of the system. This paper uses the pulse coupled neural network to extract the feature parameter for the system from the spectrogram, and make a contrast with linear prediction cepstrum coefficient and mel frequency cepstrum coefficient. The experimental results show that entropy greatly reduces the system data, accelerates the real-time and improves recognition rate of the system.This paper firstly introduces the background of the research and the present status of the speech recognition in domestic and foreign,as well as simply presents the theoretical foundation of the speech recognition and the basic knowledge of the phonetics. Subsequently, it makes detailed introduction of the pretreatment, feature parameter extraction and recognition models of the system. The improved spectral subtraction listed in this paper used to improve recognition rate in the noisy environment.Moreover, the paper introduces how to use the pulse coupled neural networks to extract the feature parameter from the spectrogram. Finally, it has realized the speaker dependent speech recognition system which is based to the proposed method.