Vioce Process System Based On FPGA

The development of integrate circuit in 20th century has provided admirable opportunity to digital audio technology.It has developed the audio technology rapidly.With the development of audio transact technology,one of its offset, voice identification technology has been widely used.However, this technology has not yet used with satisfaction.The sticking point is how to improve the veracity of the identification.Meanwhile,with the development of the technology,the design of the electronic circuit is getting rid of traditional pattern,while using FPGA to design electronic circuit is becoming the trend of design.Because using FPGA to design electronic circuit have the advantage of short time and relative fewer funds,and can make product as electronic circuit to product as CMOS chip.Based on above analysis, this thesis brings forward the usage of digital filter, which uses the distributed FIR to get excellent linear feature.By the means of search table,multiplicative tool is seldom used and resource is optimized.High SNR data is gained by filtering of filter and combined emulation of Quartus II and Matlab.After comparing with three algorithms(Linear Prediction Coefficient,Linear Predictive Cepstral Coefficients,Mel Frequency Cepstrum Coefficient)on feature extraction and two styles(Dynamic Time Warping,Hidden Markov model)on Recognition algorithm,a new design on embedded audio signal recognition,which is based on FPGA and Nios II softcore,is proposed.The scheme uses Cyclone II FPGA chip of Altera Co. and combines software and hardward design.FPGA completes pre-emphasis,windows and frames,and then Nios II fulfills endpoint detection,feature extraction and pattern recognition,finally the Matlab deals with emulation.In the end of the thesis,endpoint testing arithmetic and test method has been improved properly,which can improve the identification of the system and the veracity of the identification.