FOA-LM算法及其在语音信号稀疏分解中的应用

摘要

Signal sparse representation has important applications in many aspects of signal processing, but the problems of having large memory capacity and huge computational complexity in sparse decomposition, restrict its practical usage. Particle Swarm Optimization(PSO)algorithm and Fruit Fly Optimization Algorithm(FOA)being iterations random algo-rithms do better in finding the optimization solution with greater probability, so they can greatly improve the speed of the speech signal sparse decomposition, but they have the weakness of slow convergence in the later and less precision. Levenberg Marquardt(LM)algorithm has the advantage of the fast convergence speed and high precision, but LM algorithm depends on the initial value, which restricts its practical usage. Combining the advantages of FOA algorithms and LM algorithms, the initial parameters of Gabor atom are solved by FOA algorithm, and then the LM is used to optimize the initial parameters. The simulation results show that the solution of the optimization can be easily and quickly obtained by FOA-LM.%信号的稀疏表示在信号处理的许多方面有着重要的应用，但稀疏分解计算量十分巨大，难以产业化应用。粒子群优化（PSO）及果蝇优化（FOA）等智能算法具备前期收敛速度快，全局搜索能力强的优点，应用到语音信号的稀疏分解中，虽然大大提高了语音信号稀疏分解的速度，但是该类算法后期的收敛速度较低，稀疏分解速度仍然偏低。拉凡格氏（LM）算法具有收敛速度快，精度高的特点，但是LM算法依赖初值，这使它的应用受到了限制。结合智能算法FOA及LM算法的优点，采用FOA算法求出Gabor原子参数初值，利用这些初值进行LM迭代搜索最优原子。仿真结果表明，基于FOA优化算法和LM算法相结合的方法，具有收敛速度快，精度高的特点，有较高的实用价值。