基于改进BP神经网络的负性情绪语音识别  

Speech-oriented Negative Emotion Recognition Based on Improved BP Neural Networks

作　　者：刘忠锋 何亮[2] 

机构地区：[1]中国石油伊拉克公司,哈法亚项目伊拉克阿玛拉62001 [2]南京理工大学自动化学院,江苏南京210094

出　　处：《温州大学学报（自然科学版）》2015年第3期17-24,共8页Journal of Wenzhou University(Natural Science Edition)

摘　　要：负性情绪对于临床治疗的效果有着巨大影响.语音是人类表达情绪的主要方式之一,通过语音识别患者的情绪状态,可以帮助我们更简便、更快捷地监控病人的情绪,从而可以更快更有效地采取措施降低负面情绪带来的不良影响.对一种改进BP神经网络进行了扩展,拓展了用于情感识别的语音特征向量的冗余度,采取主成分分析方法对语音特征向量进行降维处理,并对语音样本进行去野点处理,从而使得该BP网络同时具备了对于愤怒和悲伤两种负性情绪的良好识别能力.This paper exposes that negative emotions inflicts deep impact on the effects of clinical care because speech is one of the major patterns for human beings to express emotions. We are helped to monitor the emotional states of patients faster and simplier by automatically detecting negative emotions from speech so as to take effective measures to lower the adverse effect brought by negative emotions. In addition, the paper also introduces and extends an improved back propagation (BP) network. The proposed approach expands the redundancy of the characteristic vector for emotion recognition, applies a Principal Component Analysis (PCA) algorithm to select the most influential voice attributes, and adopts a PCA-based method to reduce the effect of outliers in the sample sets. Simulation study shows that the proposed approach is capable to effectively recognize favorable recognition capability from both anger and sorrow emotions.

关 键 词：负性情绪 BP网络 前向选择算法 主成分分析 野点 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程] TP391.42[自动化与计算机技术—控制科学与工程]