Reference Independent Component Analysis Algorithm And Its Application

Independent Component Analysis (ICA) belongs to a class of Blind Signal. Traditional ICA aims to recover all independent source signals. However, one or several source signals of interest only needed in many real situations, researchers have to choose the desired independent components after calculating all the independent components, which is not only time-consuming but also unstable.Independent component analysis with reference(ICA-R) can extract the desired source signal from mixtures by incorporating priori information into the learning algorithm as reference signal. ICA-R has fast speed and eliminates the uncertainty of the traditional independent component analysis method by using little priori information.But a crucial problem to the algorithm is how to design a reference signal in advance, which should be closely related to the desired source signal. If the desired source signal is very weak in mixed signals and there is no enough priori information about it, the desired signal is difficult to recover. The priori information and its using method are diverse, the selection of the reference signal is different.In this paper, according to the mechanism and characteristic of speech signal transmission, we put forward to the speech modeling method based on Bessel function expansion, using Bessel function expansion coefficients as transform coefficients, a small amount of coefficients are utilized to build a reference signal. At the same time, the speech differences between the speech signal and many noise signals of spectral characteristics are analysed, the envelopes of the power spectral of speech are utilized to construct reference signals. ICA-R is applyed to speech enhancement by properly constructing the reference signals of target speech signal. Experimental results show that the above two methods can achieve the purpose of the speech enhancement...