基于过程神经元的语音情感识别的建模与仿真  

Modeling and simulation of speech emotional recognition based on process neural network

作　　者：叶吉祥[1] 陈晋芳[2] 

机构地区：[1]长沙理工大学计算机与通信工程学院,长沙410114 [2]中南大学信息科学与工程学院,长沙410083

出　　处：《计算机工程与应用》2014年第15期187-190,239,共5页Computer Engineering and Applications

基　　金：湖南省自然科学基金重点项目(No.10jj2050)

摘　　要：为克服由传统语音情感识别模型的缺陷导致的识别正确率不高的问题,将过程神经元网络引入到语音情感识别中来。通过提取基频、振幅、音质特征参数作为语音情感特征参数,利用小波分析去噪,主成分分析(PCA)消除冗余,用过程神经元网络对生气、高兴、悲伤和惊奇四种情感进行识别。实验结果表明,与传统的识别模型相比,使用过程神经元网络具有较好的识别效果。To improve the problem of the low recognition accuracy caused by the defect of the traditional speech emotion recognition model, this algorithm of process neural networks is introduced to the speech emotion recognition. This paper extracts the speech emotion features of fundamental frequency, amplitude, sound characteristic, and uses the method of wavelet analysis to reduce noise, the Principal Component Analysis（PCA）to reduce the redundancy, and carries on the experiment of classification and recognition of the four speech emotions of anger, happiness, sadness and surprise. The result proves that the method of process neural network has better recognition effect on the four speech emotions compared with the traditional recognition model.

关 键 词：语音情感 智能模型 神经网络 主成分分析 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]