Abstract: Hough Transform which is always used in image processing，turns the detection of the direction to the peak point detection in transform space in order to estimate the mixed matrix.This method is a new algorithm——HT-LSM algorithm.The paper firstly introduces blind mixing model recovery with Hough Transform，then combines the improved Hough Transform and least-squares method，it comes out to improve the speed and the accuracy at the same time.This method is applied to underdetermined speech signal separation，and achieves good effect of the experiment.

摘要： 利用Hough变换可以检测观测空间中的直线方向从而确定混叠矩阵的方法，提出了欠定盲源分离中估计混叠矩阵的一种新算法——HT-LSM算法。该算法在介绍欠定盲信号分离基本原理的基础上，介绍基于Hough变换的盲信道估计算法，并将改进后的Hough变换与最小二乘法相结合，在不影响检测结果速度的同时又进一步提高了检测精度，应用到欠定语音信号分离中，取得了良好的实验效果。