基于免疫猫群优化算法的矢量量化的码书设计及语音识别

发布时间：2018-05-24 02:51

本文选题：猫群优化算法 + 克隆扩增　；参考：《模式识别与人工智能》2014年07期

【摘要】：在矢量量化的码书设计过程中,针对传统的LBG算法对初始码书选取的依赖性及易陷入局部最优的缺陷,提出基于免疫猫群优化算法的矢量量化码书设计.将整个种群分为搜索组和跟踪组,运用克隆扩增算子在搜寻组中进行局部搜索,根据适应度值大小调节变异个体数目,保持解的多样性.运用动态疫苗提取与接种算子使跟踪组个体基因与疫苗进行交叉变异,向最优解靠拢,防止无监督交叉变异可能引起的退化现象.通过浓度平衡算子和选择算子更新子代种群,防止种群"早熟".将训练出全局最优码书输入到HMM模型进行训练和识别,实验结果表明,基于免疫猫群优化算法的矢量量化码书设计不依赖于初始码书选取,鲁棒性强且降低语音识别误差率.
[Abstract]:In the course of codebook design of vector quantization (VQ), a novel codebook design based on immune Cat Swarm Optimization (IQSO) is proposed to overcome the dependence of the traditional LBG algorithm on the initial codebook selection and the defect that it is easy to fall into the local optimal condition. The whole population is divided into search group and tracking group. Local search is carried out in search group by clone amplification operator, and the number of variant individuals is adjusted according to fitness value, and the diversity of solution is maintained. Dynamic vaccine extraction and inoculation operators are used to make individual genes in the tracking group cross mutate with the vaccine in order to close to the optimal solution to prevent possible degradation caused by unsupervised crossover mutation. The concentration balance operator and the selection operator are used to renew the progeny population to prevent the population from precocity. The global optimal codebook is input into the HMM model for training and recognition. The experimental results show that the vector quantization codebook design based on immune cat swarm optimization algorithm does not depend on the initial codebook selection and is robust and reduces the speech recognition error rate.
【作者单位】：天津理工大学智能计算及软件新技术重点实验室;
【基金】：国家自然科学基金项目(No.61001174) 天津市高等学科科技发展基金项目(No.20071308)资助
【分类号】：TN912.34

【相似文献】