一种改进的瞬时混合语音信号盲分离算法

An improved algorithm for blind signal separation of instantaneous speech mixtures

给出语音信号分离的数学模型,并确定了可分离的假设条件和分离准则.通过分析2种学习规则的步长变化对分离效果的影响,提出改进的学习规则.应用改进的学习规则实现自适应算法对语音信号的盲分离,消除步长递减过早或过晚的现象,收敛速度快,稳态性强,分离效果较好.

In this paper the mathematical model of signal speech separation is established,and the separable assumption conditions and separation criteria are determined.It proposes improved learning rules of varied steps through analysis of the separation effect for two variable-length learning rules,applies the improved rules in the adaptive algorithm to realize blind separation of speech signals,and eliminates the phenomenon that the step width lapses either too late or too early,so that stability of separation signals is strong with effective separation.