论文部分内容阅读
针对说话人声纹身份认证一体化设计需求,利用曲线拟合方法改进Mel滤波器组,开展声纹特征参数提取的SoC设计研究。首先选取MFCC,IMFCC,MidMFCC的Mel频率刻度上的平均值作为拟合参数,得到拟合函数;然后利用插值方法改进Mel滤波器组中各滤波器的加权中心频率;通过Matlab进行算法设计和仿真分析,采用Quartus II和Modelsim实现代码编写、编译、时序仿真和综合;最后利用SOPC Builder技术将Verilog模块封装为IP核,并下载到DE2开发板中进行验证。研究结果表明,本系统提取出的多维特征参数具有更好的鲁棒性,有利于提高识别率。
In order to meet the requirement of integrated authentication of speaker’s voiceprint authentication, the Mel filter bank is improved by the curve fitting method, and the SoC design of feature extraction of voiceprint is studied. Firstly, the average values of Mel frequency scales of MFCC, IMFCC and MidMFCC are selected as the fitting parameters to obtain the fitting function. Then, the interpolation method is used to improve the weighted center frequency of each filter in the Mel filter bank. The algorithm design and simulation Analysis, using Quartus II and Modelsim code writing, compilation, timing simulation and synthesis; Finally, use SOPC Builder technology to Verilog module is packaged as IP core, and downloaded to the DE2 development board for verification. The research results show that the multi-dimensional feature parameters extracted by this system have better robustness and are helpful to improve the recognition rate.