一种遗传算法的声带模型参数反演方法

A Parameter Inversion Method in Vocal Fold Model by Improved Genetic Algorithm

采用遗传算法进行声带模型参数反演能够有效地实现概率意义的全局搜索,但存在易陷入局部最优解和收敛性能弱等缺陷,针对这些问题本文提出了一种改进遗传算法的声带模型参数反演方法。通过声门波形参数构建目标函数,引入精英保留策略,选择适应度高的精英个体保留至下一代种群,对普通种群进行双向邻域搜索;通过目标函数值进行模型参数反演操作,得到模型参数最优解。实验结果表明,反演后得到的特征相对误差不超过1.5%,改进后的加权平均相对误差比改进前减小了0.11%。说明改进后的遗传算法搜索性能好,收敛性能高;得到的特征参数精确度更高。

When Genetic Algorithm is used to inverse the parameter of vocal fold model,it is easy to get into the local optimization solution and will show a slow convergence performance.For these problem,a parameter inversion method in vocal fold model is proposed by improved genetic algorithm.The objective function is constructed by using glottal waveform parameters and the highly adaptable elite individuals were selected to retain to the next generation by elite retention strategy.For the general population,bidirectional neighborhood search is carried out.With the value of objective function,the optimal solution of model parameters was obtained by inverting.Experi-mental results show that the obtained characteristic relative error after inversion is no more than 1.5%and the modified weighted average relative error is 0.11%lower than that before the improvement.It shows that the improved genetic algorithm has good search performance and high convergence.The obtained characteristic para-meters are more accurate.