Isolated Word Speech Recognition System Design

Speech recognition technology is adopted for the identification and understanding of the machinery voice signal into the text or order high-tech, over the last 20 years, speech recognition technology has made remarkable progress, beginning from the laboratory to the market. With VLSI technology, speech recognition technology has entered a post-PC era research direction, embedded speech recognition chip research has become a major direction.In this paper, the speech recognition system for a basic introduction, and discussed a based on the small isolated word speech recognition system hardware to achieve. And on the basis features of preemphasis modules and feature extraction module suggested a more appropriate hardware to achieve algorithms improved. And providing a framework structure for the realization of the hardware. In processing modules, in order to simplify the difficulty of hardware design and improve the accuracy of identification from the two done, first, to take a similar pre heavier modules algorithms, and in the calculation of the division of energy use rolling methods, and successfully avoided France on the design;Second, the end of the algorithm to improve the detection, long and dynamic windows were used to zero threshold from poor judgment algorithms. From the experimental results of comparative results satisfactory. In the extraction of features, discards the design commonly used in software since the relevant equation Dubin algorithms combined approach, and a more appropriate hardware design era Ditui algorithms. Further optimized hardware design.The key element is the realization pattern matching modules. According to system requirements, design a pattern matching algorithms based on DTW, as is based on isolated words and the code is small, so using DTW algorithm is a more ideal choice, but there are two DTW algorithm greatest mistake : First, the sensitivity to the end;Second, the operation of the larger. In this issue we take a compromise, in the traditional accelerated algorithms for improved with the introduction ofsemi-relaxed style at the end. To do so would have two advantages : First, it DTW algorithms to significantly improve the extreme sensitivity;Second is to improve the semi-liberal style better suited hardware at the end achieved. From the experimental results of achieving better results..