The Realization Of Speaker-Recogniaztion Embeded System Based On FPGA

In recent years, speaker recognition is a rapid developing technology. To let the machine recognitize speaker by speech is the main purpose of this research. With the growing of computer and electronic technology and the improving of the theory of speaker recognition, speaker recognition technology used in real life and work demands more and more pressing. Pass through the research from the 1930s to date, many excellent speaker recognition models have been built.SOPC is a system on programmable chip based on FPGA. System based on SOPC should first reasonable planning system structure, choose algorithm and peripherals which are suitable for embedded platform, and make full use of limited resources for the realization of system functions.The main task of this issue is under systematically consider the resources and effect of recognition to choose suitable algorithm and speech parameters, and to realize the embedded speaker recognition system based on SOPC which has certain precision and speed.The theoretic basic of this topic is the techinique about speaker recognition,signal processing and embedded SOPC.This topic put forward a set of feasible systems architecture.After do thorough research on various algorithm in the speaker recognition theory and the characteristic of system,this topic carried on the reasonable choice and optimization to the algorithm.With consumes the time situation and according to the algrithm characteristic to choose realize the system by software or unify software and hareware module.The content of this paper mainly includes following several aspects:1. Thoroughly research the pitch detect and speech feature extraction algorithm.Consider the system characteristic, this topic design reasonable sound acquisition and speech pitch detect flow,and in view of the speech feature extraction withdraw a processing flow base on Dual-CPU SOPC.The experiment show that these designs greatly enhanced the speed of system with low error rate and good reliability.2. This topic do research on speaker recognition algorithm based on GA-K-VQ (genetic algorithms-K-means clustering vector quantization) and analyze the recognition algorithm with emphasis, and finally realizes on the SOPC system. Through to the algorithm improvement, design approprate hardware acceleration module, effectively enhanced the speaker codebook establishment efficiency. Test shows that without affect on the rate of identification premise, the hardware realization of code book building, which is the longest time-consuming process, may quicker 2-4 orders of magnitude than software.3. This topic realize the design proposal on hardware, and design essential man-machine interactive connection, finally realized the speaker recognition system with excellent recoginize rate and short recoginition time. The experiment indicated that the system architeture proposed by this paper is efffective and feasible. Speaker recoginition system based on SOPC has unique superiority in the speed, precision and expandbility, The developable space of this system is broad.