摘要
传统的说话人识别方法在识别能力和稳定性上存在不足,对各种方法没有较好的评价标准。提出了基于平均距离和相关性的评价准则,在比较应用MEL滤波器组和小波包分解提取特征参数的基础上,提出了混合特征参数的提取方法。实验结果表明,距离和相关性准则可以较好的评价各种参数模式的分类能力,在无关的说话人识别中,混合参数在分类能力和抗噪性方面优于传统方法。
Traditional speaker recognition has some shortcomings in recognition capability and stability. There is no estimating rule in speaker recognition. At first, the theory of synthetical formula based on distance and relativity was advanced. Than, beaded on comparing different parameters of MEL filter group and wavelet packet transform, the mixed parameter was advanced. Simulation results show that distance and relativity formula are highly successful in estimating recognition capability of different method. The mixed parameter method is efficient in classifying and anti-noise for the Text Independent Speaker Recognition than traditional methods.
出处
《系统仿真学报》
CAS
CSCD
北大核心
2008年第4期926-930,共5页
Journal of System Simulation
关键词
说话人识别
混合参数
距离和相关性准则
MEL滤波器组
小波包
speaker recognition
mixed parameter
distance and relativity formula
MEL filter group
wavelet packet