论文部分内容阅读
在动态声纹密码认证系统中,针对声纹与文本得分归一化时出现样本类内散度和样本间距离相矛盾的问题,文章设计了一种基于方差与协方差调节因子的归一化方法,从缩小样本类内离散度和增大样本间距离两个方面进行归一化研究。实验结果表明:引入方差和协方差调节因子的组合归一化方法,在样本类内散度基本不变的情况下,有效增加了样本间的距离,通过支持向量机分类器进行分类,准确率提升了2.02%。
In the dynamic voiceprint password authentication system, aiming at the contradiction between the divergence of samples and the distance between samples when voiceprint and text score are normalized, a normalization based on variance and covariance adjustment factor is designed. Method, the normalization is studied from the aspects of reducing the degree of dispersion in samples and increasing the distance between samples. The experimental results show that the combination normalization method of variance and covariance adjustment factor is introduced. Under the condition that the divergence in samples is almost the same, the distance between samples is effectively increased, and the classification is made by SVM classifier. The accuracy Increased by 2.02%.