Research Of Speech Separation Based On Improved ICA Algorithm

Blind source separation is the process which the source signal is recovered from the observed signal based on the statistical characteristics of the source signal under the condition of the source signal and the channel transmission model unknown.In recent years,blind source separation has become a hot research topic in the field of signal processing and artificial intelligence.In the blind source separation algorithm,the independent component analysis(ICA)algorithm is widely studied and applied due to the less requirement of the application environment.In this paper,two improved algorithms are proposed and applied to speech signal separation based on the research of ICA algorithm.An improved FastICA algorithm based on negentropy is proposed.This method uses the negative negentropy as the cost functions of FastICA algorithm.Furthermore,the conjugate gradient algorithm takes the place of the Newton iteration method to solve cost function,due to the fact that the conjugate gradient algorithm has low computational complexity and good stability.This improved algorithm can effectively solve the problem of high computational complexity and sensitivity to the initial value caused by Newton method.Eventually,the effectiveness of the improved algorithm is verified by simulation experiment from three aspects:run time,iteration number and PI performance index.The traditional natural gradient algorithm adopts a fixed step-size factor,which cannot solve the contradiction between the convergence speed and the separation effect.Variable step-size natural gradient algorithm can dynamically adjust the step size factor,but often cannot effectively track the degree of signal separation.In view of the above problems,this paper presents a new adaptive natural gradient algorithm.The algorithm uses the similarity among the separated signals to define the separation state of the signal,and controls the selection of the step factor.At the same time,the momentum term is introduced in the algorithm to further improve the convergence speed of the algorithm.The experimental results show that the adaptive blind source separation algorithm proposed in this paper has better separation performance.