Study And Design Of Isolated-word Speech Recognition System Based On SOPC

As the speech recognition technology getting mature, it has been applied to industry,military, traffic, civilian, etc. Furthermore, the application areas of speech recognition will beexpanded increasingly with the development of the speech recognition technology.Compared with large vocabulary speech recognition systems based on PC platform, thespeech recognition module on the embedded system is required to consume less systemresources and to be simplified to use. And it could be used as independent speech recognitionsystem or the speech recognition module of larger systems. In order to meet the requirementsfor some system, we achieved a simple isolated-word speech recognition system, and thissystem is able to recognize isolated-words which is two or three Chinese characters(such as“open”,“stop”,“close”, etc). According to speech recognition system’s accuracy andreal-time demands, we have finished the following research works. Firstly, as the Endpointdetection has significant impact on the accuracy of speech recognition system, this systemseeks Endpoint detection algorithm which is suitable for the the SOPC systems.So all this willimprove the accuracy of speech recognition system. Secondly, the Pattern Matching is basedon the Dynamic Time Wrapping algorithm. During Pattern Matching，get different values ofdistance according to several different Local Decision Functions,and then get the average ofthe different values of distance, and that is the the end result.This method improved thereliability of the speech recognition system.SOPC (System on Programmable Chip) is formulated by Co. Altera. And it is andesigning method for the embedded system based on FPGA. The SOPC technology combinesthe designing process of software and the designing process of hardware ware.And is easy toaccomplish. This speech recognition system uses EP2C70FPGA chip as the main part of thesystem.The result of the experiment shows that the speech recognition system works well andit can satisfy the demands of the simple isolated-word speech recognition systems.