Design And Research Of Speech Blind Separation System Based On DSP

This article studied the FastICA algorithm first,then we designed and implemented a speech blind separation system based on DSP,which laid a solid foundation for the speech blind separation in the application of embedded system.The blind source separation(BSS)refers to the process of restoring each independent source signal from the observed signals just relying on the statistical properties and some priori conditions of source signals,in the situation of not knowing the source signals and the transmission channel parameters.With the development of computer technology,people has higher requirement for signal processing and its application field is becoming more and more broad,at present BSS has been widely used in speech detection,communication,biomedical engineering and earthquake,etc.As a kind of super-Gaussian signals,the blind separation technology of speech signal is a current hot research area.Now solving the problem of blind source separation is mainly using independent component analysis(ICA),and this article studies the ICA in the application of linear instantaneous without noise mixed voice model.In the study of information theory,researchers have designed a lot of practical algorithm based on the maximum likelihood function,the minimum mutual information criterion and the maximum entropy,such as the maximum likelihood estimation algorithm,FastICA algorithm,Infomax algorithm,etc.This article selects FastICA algorithm,which is based on maximizing the non-gaussianity of source signals.FastICA gets the separation matrix through repeated iterations,and then extracts the independent source signals,and it's widely used for its high speed of calculation.Besides,with the development of the microprocessor technology,the performance of digital signal processor has improved a lot,which provides the ideal hardware platform for the practical application of blind source separation algorithm.This article first simulates the FastICA algorithm by MATLAB,proving the feasibility of its practice in the two-way mixed speech blind separation,then transplanting the algorithm to the TMS320C6748 evaluation module,which is a floating-point DSP development board,and then achieving a two-source-signal and two-channel blind source separation system.Then this article puts forward two kinds of blind source separation scheme,the online speech separation and offline speech separation,solving the real-time problem and the order uncertainty of separation results.Finally,the experiments prove that the blind source separation system achieves good separation effect.