摘要
针对语音情感的动态特性,利用动态递归Elman神经网络实现语音情感识别系统。通过连接记忆上时刻状态与当前网络一并输入,实现Elman网络模型的状态反馈。基于此设计了语音情感识别系统,该系统能在后台修改网络类型,并实现单语句与批量语句识别模式。针对系统进行语音情感识别实验表明,基于Elman神经网络的语音情感识别在同等参数模型设置前提下优于BP神经网络识别效果,且BP神经网络参数设置较Elman网络敏感。
This paper utilized the Elman dynamic recurrent neutral network to realize speech emotion recognition system for dynamic characteristics of speech.Meanwhile,it realized state feedback of Elman neutral network through input both of connection memory from last state and current state together.Finally,it designed speech emotion recognition system which could not only modify network structure types in backstage,but also practiced two mode of signal speech recognition and batch proces-sing recognition.Based on this platform,the experiments show that the recognition effect of Elman is superior to BP network un-der the same model parameters.Furthermore,parameters setting of BP are more sensitive than Elman network.
出处
《计算机应用研究》
CSCD
北大核心
2012年第5期1809-1814,共6页
Application Research of Computers
基金
中央高校基本科研业务费专项资金资助项目(2012QNZT060)
第49批中国博士后科学基金面上资助项目(20110491272)
中南大学博士后基金资助项目(2010-2012)
湖南省教育厅青年基金项目(11B070)
关键词
语音情感识别
ELMAN网络
BP网络
MFCC
speech emotion recognition
Elman network
back propagation network
mel frequency cepstrum coefficient