基于三二次优化的语音卷积盲分离方法

Blind Convolutive Separation Algorithm for Speech Signals via Tri-Quadratic Minimization

针对语音信号卷积混迭模型,利用语音信号近似独立及短时平稳特性,提出一种基于三二次迭代优化的混迭矩阵多分量联合估计方法,交替估计三组待定参数的二次代价函数实现联合块对角化,在时域解决卷积混迭盲分离问题.理论分析了所提方法的计算复杂度和迭代收敛性.实验表明,相比于类Jacobi方法,所提方法收敛速度更快,且全局拒噪水平平均改善5dB,巴克谱失真系数改善达0.12.

A convolutive blind source separation algorithm based on tri-quadratic joint block diagonalizafion is proposed by utilizing the mutual-independence and quasistationarity properties of speech signals. The three subgroups of the undetermined parameters are iteratively estimated through minimizing the corresponding quadratic cost subfunctions alternatively. Analyses prove that the proposed algorithm has low computational complexity and robust convergence. Compared with the classical Jacobi-like methods, simulation results show that the proposed algorithm can improve both the global rejection level about 5dB and the Bark spectral distortion factor about 0.12 in average with faster convergence rate.