The Design And Realization Of Sppech Recognition Algorithms Based On Embedded Platform

Nowadays the speech recognition systems have achieved better results in the laboratory environment, however there still have some problems when it be applied into the embedded devices , the problems includes the slow calculating speeds, the limited capacity and the restrict response time and etc. So it seems that we now still have long way to go to bring the technology into application, and under such circumstance it is meaningful to focus our energy on the research and realization of speech recognition algorithms which based on embedded system.This paper mainly discussed the isolated words speech recognition technology, and the core task of the paper is to design and accomplish speech recognition algorithms based on the embedded system.Firstly we analyzed the Mel frequency cestrum coefficient extraction algorithms, according to the principals of auditory model of human ear, MFCC algorithms transport the linear frequency spectrum into Mel non-linear frequency spectrum, and through logarithm and discrete cosine transport calculation it gives the output values, and we have achieved better results when adopt energy and dynamic differential coefficients into research.Secondly we mainly analyzed Hidden Markov Models , it consist of three different parts: a. "Forward-Backward algorithm" try to give the probability of observe sequence; b. Viterbi algonthm try to seek the optimized state transition path; c. Baum-Welch algorithm try to find optimized the training parameters, and based on the experiment results we confirm that Hidden Markov Models can be well described the speech signals.Thirdly we described the structure of OMAP5912 hardware platform and peripherals interface.OMAP5912 hardware platform is Taxes Instrument's products which is target for the portable date terminal products. It has combined high performance TMS320C55xDSP kernel and low power ARM926EJS kernel in a single SoC chip, to meet the needs of the development for embedded speech recognition system. Also in this paper we discussed the process of building an embedded system environment, it makes up of three sections: a. develop the universal boot loader program on the embedded system; b. transplant Monta Vista Linux operation system on the embedded device ; c. design JFFS2 file system.In the end we mainly discussed embedded graphic user interface designing works, we now have finished the compile and transplant of MiniGUI program codes, realized the development of the graphical interfaces application program based on the MiniGUI-Lite mode ,and successful completed speech recognition graphical interfaces with command line control functions. After the whole work we now have completed the tasks of building an embedded speech recognition system successfully.