Research And Realization Based On DSP Of Little Voice-identification-Authentication System

With the popularity of small portable devices,a simple and effective way to authenticateidentity is required urgently to ensure the safety and individuation of devices.This applicationoccasion calls for simplicity and low cost on the basis that recognition rate is guaranteed. Aimedat this application, this paper realizes a little,text-dependent speaker recognition system of closedset to achieve identity authentication.The main procedure of speaker recognition is split into pretreatment, feature extraction andpattern matching. In this paper we briefly compare all kinds of algorithm principles andadvantages and disadvantages in entire implementation process; then choose FRED endpointdetection algorithm, MFCC, vector quantization(VQ) and minimum distance judgment methodas main algorithms to design this system, and provide detailed theory statement and algorithmflow design.The hardware platform of this system is based on an ICETEK-VC5502-AE evaluationboard. In this paper we sketch out the features of the TMS320VC5502DSP chip, as well asinterface principles that the DSP chip is interfaced with the stereo audio codec chip, the SDRAMchip and the FLASH chip; then fulfill the software algorithm based on CCS3.3IDE; in addition,we carry through code optimization.