Dual-channel speech enhancement based on stochastic optimization strategies

In this paper, we propose an improved stochastic optimization algorithm called Learning-based Particle Swarm Optimization (LPSO) to design adaptive filter for dual-channel speech enhancement application. The novel algorithm employs a multi-swarm model based on knowledge learning method and dynamic search of global best (gbest) technique, to improve the performance of the Standard Particle Swarm Optimization (SPSO). The knowledge learning method uses the knowledge obtained in the searching process, and the dynamic search of gbest technique simulates the act of human randomized search behavior. The proposed algorithm shows an outstanding performance in dual-channel speech enhancement, and outperforms the SPSO, genetic algorithm (GA), and Normalized Least Mean Squares (NLMS) in a sense of stability and SNR-improvement.