混合蛙跳算法神经网络及其在语音情感识别中的应用  

Shuffled Frog-leaping Algorithm Based Neural Network and Its Application in Speech Emotion Recognition

作　　者：余华[1,2] 黄程韦[1] 张潇丹[1] 金赟[1] 赵力[1] 

机构地区：[1]东南大学水声信号处理教育部重点实验室,江苏南京210096 [2]南京信息职业技术学院电子信息学院,江苏南京210013

出　　处：《南京理工大学学报》2011年第5期659-663,共5页Journal of Nanjing University of Science and Technology

基　　金：国家自然科学基金(60472058;60975017;51075068);江苏省自然科学基金(BK2008291)

摘　　要：该文将混合蛙跳算法(SELA)优化方法应用于人工神经网络训练中,对6种语音情感进行了语音情感特征的分析与识别。研究了谐波噪声比特征随情感类别的变化特性。利用混合蛙跳算法训练随机产生的初始数据优化神经网络的连接权值,快速实现了网络收敛。实验比较了BP神经网络、RBF神经网络和SFLA神经网络的语音情感识别性能。结果表明,SFLA神经网络的平均识别率分别高于BP神经网络和RBF神经网络4.7%和4.3%。The shuffled frog-leaping algorithm（SFLA） is applied to the speech emotion recognition in neural network training.The freatures of the six speech emotions are extracted and recognized.The changes of harmonics-to-noise ratio（HNR） features with different emotions are studied.The random initial data trained by the SFLA is used to optimize the connection weights and thresholds of the neural network,and the network can converge fast.The recognition capability of the BP,RBF and SFLA neural networks are compared experimentally.The results show that the recognition ratio of the SFLA neural network is 4.7% better than that of BP neural network and 4.3% better than that of the RBF neural network.

关 键 词：混合蛙跳算法 神经网络 语音情感识别 辨识率 

分 类 号：TP273[自动化与计算机技术—检测技术与自动化装置]