Blind Source Separation Based on Improved Particle Swarm Optimization

Blind Source Separation (BSS) is a recently addressed speech signal processing method, the traditional searching scheme use gradient-based algorithm; However, the convergence of it often depends on choosing of a learning step, it couldn´t resolve the problem of lower velocity of convergence. To overcome the drawback, an efficient BSS algorithm based on improved Particle Swarm Optimization (PSO) is presented. We introduce evolution speed and aggregation degree to update dynamic inertia weight in PSO. Then define fitness function of PSO based on BSS. Finally, the detail algorithm of BSS is presented. Experimental results on mixed voice signal indicate that the established algorithm of PSO can quickly and effectively get optimal resolution to BSS.