PSO与PCA融合优化核极限学习机说话人识别算法仿真

Algorithmic Research on Kernel Extreme Learning Machine for Speaker Recognition Based on PSO and PCA Optimization

基于机器学习理论开展说话人识别的研究取得了很大进展,在基于核极限学习机(kernel extreme learning machine,KELM)和梅尔倒谱系数(mel-frequency cepstral coefficients,MFCC)说话人识别研究基础上,通过主成分分析算法(principal component analysis,PCA)对MFCC进行降维优化、粒子群优化算法(particle swarm optimization,PSO)对KELM初始输入参数进行优化开展基于PSO和PCA融合优化KELM说话人识别算法研究。改进后的算法在MATLAB平台上仿真通过,并与MATLAB语音工具箱提供的神经网络和支持向量机说话人识别算法做了性能对比分析。仿真研究结果表明:通过PSO和PCA融合优化改进的KELM,初始输入参数可以任意确定并且不需要迭代更新,并能有效克服因初始权重随机确定导致的性能不稳定,进一步提高分类匹配和运算速度,具有很好的推广应用价值。

Great progress has been made in the research of speaker recognition based on machine learning.Based on the study of speaker recognition by kernel extereme leaming machine (KELM) and MFCC to start a study on particle swarm optimi zation (PSO) and principal component analysis (PCA) optimized KELM speaker recognition algorithms,the MFCC was optimized by PCA and the KELM was optimized by PSO.The improved algorithm was simulated on MATLAB and the performance comparative is made with BP and SVM.The simulation result shows that PSO-KELM can arbitrary initial input parameters and does not require iteration update and it can effectively overcome the unstable performance due to the random determination of the initial weight,further improve the accuracy of classification matching and have a good application value.