Speech Endpoint Detection And Recognition Based On Improved Algorithms

Although the research of speech recognition system has lasted for nearly 50 years,the existing one still has many shortcomings,such as time consuming,high cost,poor anti-noise ability and narrow application scope.But the practical speech recognition system requires real time and high accurate capacities in the complex daily environment.Therefore,the development of endpoint detection algorithm and fast recognition algorithm has become an important topic in the research of speech recognition.In this paper,a lot has been done about the speech endpoint detection algorithm based on wavelet packet transform and high-order cumulant and speech recognition algorithm based on adaptive immune clonal neural network.They are summarized as follows:First,the basic steps of speech recognition are introduced,and the current development of it is reviewed.The basic knowledge of preprocessing,feature extraction,recognition and training is also described.Second,an improved endpoint detection algorithm is devised based on wavelet packet and high order cumulant for the inherent lack of good robustness in the tradition ones,overcoming the weak ability in processing noise and high frequency part of speech signal due to wavelet transform.The results show that this algorithm improves the accuracy of speech endpoint detection.Then,in order to perfect the poor accuracy and weak robustness of traditional immune neural network speech recognition system,an immune clonal neural network is proposed based on an adaptive mutation probability function.A speech recognition system based on this algorithm is designed in order to improve its accuracy,aiming at isolated word recognition tasks in real time.The results indicate that the speech recognition system based on this algorithm has higher recognition accuracy and robustness.Finally,the work done in this paper is summarized,and the direction to get further studied is also pointed out.