Recognition System Of Speaker Designed By The Technology Of FPGA

Speaker recognition are emitted by the analysis of speaker's voice and the extraction of characteristic parameters, To determine whether the speaker is in the registration of the template library or not automatically, as well as,to verify the identity of the speaker. At present, Rely on its unique advantage, the speaker recognition has great application value and broad market prospect. The feature of repeat programming for FPGA is helpful for the future system's maintenance and upgrade. In addition, because of its high function of parallel data processing, FPGA can meet the requirements of market on the system real-time performance. Therefore, in this thesis using the FPGA technology to realize the speaker recognition system hardware design.First of all, in this paper, I study the speaker recognition algorithm, mainly including the speech siginal processing, the extraction of MFCC feature parameter, the recognition of VQ pattern. Based on the research of traditional algorithms, this paper adopted with short time average magnitude instead of short-time energy in the endpoint detection algorithm. Therefore, the effictive segment are intercepted more accurately. At the same time, the method of extraction of MFCC feature parameter based on the theory of stochastic resonance, effectively improve the recognition rate of the system. Speaker recognition algorithm is researched on Matlab platform, the simulation results stored in SD card, to compare with the Modelsim simulation results faciliatly.The program control module---NOIS II is designed, according to the characteristics of the SOPC system, all implementation algorithms and interface of each module are integrated into the SOPC system, bus connection relations set up in the SOPC Bulider, memory space and start address are configured, when the power is on, the program run into SDRAM. Twenty individual voice is collected, a training template library and speaker test library are setted respectively, select five hardware identification system as a reference template from the twenty template library. And then to simulate the test of 20 speech recognition through Modelsim, and compare the results of Modelsim simulation with Matlab simulation results, from the contrast of the results, although there is a small error but basically negligible, thus the function of the speaker recognition system is verified, it realizes the feasibility of speaker recognition system.The recognition rate of a speaker recognition system is very important. Although a set of real-time speaker recognition system based on FPGA is realised in this paper, but the system recognition rate is not high, future research and exploration are still needed. For example, MFCC feature extracting parameters is more effective in a speaker recognition system, but the influence of noice environment on the recognition rate is big, and we have a lot of methods to choose when choosing the characteristic vector, in order to improve the recognition rate of the system, different combination of characteristic parameters can choose or not is still needs futher research.