论文部分内容阅读
基于全变量因子分析和概率线性区分性分析的算法是目前与文本无关的说话人确认的主流算法。该文将全变量分析和支持向量机结合起来,把低维的全变量因子作为支持向量机的输入特征,并采用余弦核函数来增强低维特征的区分性,该系统取得了与当前主流算法相当的性能;进一步,将此系统得分和概率线性鉴别分析系统得分融合起来可以取得明显的性能提升。在NIST 2012说话人评测通用测试条件的女声部分,融合后的系统在情境一和三的检测代价函数相对最好的单系统分别下降了25.1%和25.2%。
The algorithm based on total variable factor analysis and probabilistic linear discriminant analysis is the mainstream algorithm of speaker-independent text-independent verification. In this paper, full variable analysis and support vector machine are combined to make the low-dimensional full variable factor as the input feature of SVM, and the cosine kernel function is used to enhance the differentiation of low-dimensional features. Equivalent performance; further, the system score and the probability of linear discriminant analysis system scores together can achieve significant performance improvements. In the female part of the NIST 2012 speaker evaluation common test condition, the combined system decreased by 25.1% and 25.2%, respectively, relative to the best single system in scenarios one and three.