偏亚高斯语音信号有效分离ICA方法研究
Method Research on Effictive Seperated ICA Algorithm to Sub-Gaussian Distribution Audio Signal
-
摘要: 针对扩展Infomax语音分离算法仅只对无偏信号的概率密度分布进行建模的不足,提出了一种有效分离的有偏亚高斯信号ICA算法。通过修改扩展Informix算法所基于的Pearson混合模型,使修改后的模型既能较好地逼近对称的概率密度分布,又能逼近非对称的概率密度分布,从而在源信号是非对称分布的情况下,能获得更好的分离质量和较快的收敛速度。Abstract: An effictive seperated independent component analysis (ICA) algorithm is obtained by introducing a skewness-adjusting parameter to the Pearson mixture density model in extended Infomax algorithm. This model with skewness-adjusting parameter can cover a wider range of sub-Gaussian distribution including asymmetrical and multi-modal ones, resulting in more precisely approximating source's density. When dealing with non-skewed mixed sources, the new algorithm can achieve less steady-state error while maintaining fast convergence speed.