论文部分内容阅读
为提高文本无关说话人识别系统的识别率,提出一种基于总体最小二乘法的无用分量投影算法.利用总体最小二乘法估计的隐含变量考虑无用分量投影矩阵的扰动,并将该扰动最小化,使基于该隐含变量求得的投影矩阵能更好地刻画无用分量空间.在美国国家标准技术署于2008年公布说话人识别数据库上的实验结果验证该方法的有效性.
In order to improve the recognition rate of text-independent speaker recognition system, an unwanted component projection algorithm based on global least squares method is proposed. The implicit variables estimated by the global least squares method are used to consider the disturbances of the unwanted component projection matrix and minimize the disturbance So that the projection matrix based on this implicit variable can better describe the space of unwanted components.Experimental results on the speaker identification database released by the National Standards Institute of the United States in 2008 verify the effectiveness of this method.