关键词: 语音盲分离, 回溯搜索优化算法, 卷积混合, 独立矢量分析, Givens旋转变换

Abstract: Aiming to overcome the limitation of initial separation matrix selection in Independent Vector Analysis（IVA）, a convolutive blind speech separation algorithm based on backtracking search optimization is proposed. The sum of complex kurtosis of separated signals in each frequency point from IVA is used as the objective function. The Backtracking Search Optimization Algorithm（BSA） is used to adjust the initial separation matrix for better separation. In the separation process, complex Givens rotation transformation is used to transform separation matrix to rotation angle for reducing the coding dimension of BSA and the difficulty of optimization decreases. The blind speech separation experiments for convolutive mixture signals indicate that the proposed algorithm performs excellent separation results and the separation property is better than the basic IVA algorithm.

Key words: speech blind separation, backtracking search optimization algorithm, convolutive mixture, independent vector analysis, Givens rotation transformation