ICA和线性神经网络在带噪语音分离中的研究

用基于独立分量分析（ICA）的盲源分离方法对强噪声背景下的混合语音信号进行分离时，如果忽略噪声的影响则会产生很差的分离效果。为克服此不足，结合噪声对消和盲源分离，提出了一种在强噪声背景环境下的混合语音分离方法，即先将带噪观测信号通过线性神经网络构成自适应噪声对消器，然后采用ICA进行分离，与增加一路噪声作为源信号的分离方法相比，该方法具有更好的分离效果。

Study of noised speech separation based on ICA and linear neural network

The neglect of the noise generally causes worse effect in the blind separation of mixed speech signal by Independent Component Analysis（ICA）,which is sampled in the environmental of strong noise.In order to overcome this deficiency,here,by means of combining the noise canceller and ICA,a new separation method of mixed speech signal under the condition of strong noise is proposed.Firstly,making the mixed signal pass through the noise canceller,which is accomplished by adaptive linear neural network,then using the ICA algorithm to separate the mixed signal.Compared with another separation method,which increasing the noise as a source signal,the proposed method has a better separation effect.