基于改进的萤火虫优化算法的混合语音盲分离

针对传统盲源分离优化算法对分离性能影响较大的局限性，提出了一种基于改进的萤火虫优化的混合语音盲分离算法。将萤火虫的飞行跨度由固定取值变为由新构造的函数自适应调整，在加快收敛速度的同时避免算法早熟现象的发生。实验结果表明，与基于自然梯度、标准萤火虫和粒子群优化的盲分离算法相比，新算法对混合语音信号的分离效果较好，在收敛速度和分离能力方面都有所提升。

Blind separation of speech mixtures based on improved glowworm swarm optimization

Aiming at the limitation of traditional optimization algorithms for blind source separation(BSS), this paper proposed a blind speech separation algorithm based on the improved glowworm swarm optimization(IGSO). The new algorithm constructed a right function to adjust adaptively the flight span, so it could not only accelerate the convergence speed but also avoid the premature convergence. The mixed speech separation experiments show that the proposed algorithm performs better than BSS based on nature gradient algorithm(NGA), GSO and particle swarm optimization(PSO), and improves both convergence rate and separation ability.