论文部分内容阅读
针对目前广泛应用于说话人识别领域的MFCC特征参数包含较少说话人特征信息的问题和SVM分类器选择核函数时受到Mercer准则限制的问题,提出了一种将混沌粒子群算法(CPSO)与核匹配追踪算法(KMP)相结合的说话人识别方法.首先通过CPSO聚类算法将MFCC特征参数进行变换处理,得到精简的MFCC特征参数(SMFCC),然后利用KMP算法对核函数的形式没有任何限制的特性和良好的分类识别性能,对约简后的SMFCC特征参数进行分类训练和识别.仿真实验结果表明,基于CPSO-KMP说话人识别方法相比主流的GMM-UBM方法,在EER性能上相对提高了31%.
Aiming at the problem that the MFCC feature parameter which is widely used in the field of speaker recognition contains less speaker characteristic information and the restriction of Mercer criterion when the SVM classifier selects the kernel function, a method is proposed to combine the chaos particle swarm optimization (CPSO) Kernel matching pursuit algorithm (KMP) .Firstly, the MFCC feature parameters are transformed by CPSO clustering algorithm to obtain a reduced MFCC feature parameter (SMFCC), and then the KMP algorithm is used to perform kernel function without any Limited features and good classification and recognition performance, the reduced SMFCC feature parameters are classified training and identification.The simulation results show that compared with the mainstream GMM-UBM method based on CPSO-KMP speaker recognition method, in the EER performance Relatively increased by 31%.