论文部分内容阅读
为了进一步提高i-vector说话人识别系统的性能,该文提出了一种鉴别性i-vector局部距离保持映射(discriminant i-vector local distance preserving projection,DIVLDPP)的流形学习算法。该算法以i-vector间的Euclid距离作为度量准则,并以最小化同类点间距离同时最大化异类近邻点间距离的鉴别性准则作为优化目标函数,利用求解广义特征值的方法,得到最终的投影映射矩阵。在美国国家标准技术局2008年说话人识别核心数据集上的实验结果表明:该算法可以明显提高目前i-vector说话人识别系统的性能。
In order to further improve the performance of i-vector speaker recognition system, a manifold learning algorithm based on discriminative i-vector local distance preserving projection (DIVLDPP) is proposed. This algorithm takes the Euclid distance between i-vectors as the metric, and takes the discriminant criterion that minimizes the distance between similar points and maximizes the distance between heterogeneous neighbors as the objective function. By using the method of solving generalized eigenvalues, the final Projection mapping matrix. Experimental results on the 2008 National Institute of Standards and Technology’s speaker verification core data set show that this algorithm can significantly improve the performance of current i-vector speaker recognition systems.