基于贝叶斯网络的语音情感识别

发布时间：2018-03-13 08:30

本文选题：语音情感识别　切入点：贝叶斯网络　出处：《华南理工大学》2014年硕士论文　论文类型：学位论文

【摘要】：情感识别的目的是让计算机能读懂人的情感，进而表达出情感，使实现能理解和表达人类情感的机器成为可能。计算机的语音情感识别能力是计算机情感智能的重要组成部分，是实现自然人机交互界面的关键前提，具有很大的研究价值和应用价值。目前，语音情感识别研究工作主要集中在语音信号预处理、语音情感特征提取、特征向量降维、语音情感识别算法等方面。贝叶斯网络是1988年由Pearl提出的一种在贝叶斯决策方法的基础上发展起来的一种统计推断方法。贝叶斯网络以独特的不确定性知识表达形式、丰富的概率表达能力、综合先验知识的增量学习特性成为近几年来理论研究的热点，，被广泛应用于辅助智能决策、模式识别、医疗诊断等领域。本文选取语音情感识别算法作为重点的研究方向，提出了将基本分类器结合贝叶斯网络应用到语音情感识别中。通过对贝叶斯网络进行了深入的研究，提出使用贝叶斯网络对基本分类器分类结果进行调整，并且提出了基于混淆矩阵的贝叶斯网络生成算法，并将本文改进的算法应用到语音情感识别的研究中，分别使用支持向量机SVM和朴素贝叶斯NB为基本分类器，使用贝叶斯网络调整分类结果，在语音情感数据库（Berlin、CASIA、SAVEE）做了实验，获得比原来算法更好的分类准确率。
[Abstract]:Emotion recognition is to let computer can understand people's feelings, and express feelings, make to understand and express human emotion machine possible. Speech emotion recognition ability of the computer is an important part of emotional intelligence, is a key prerequisite to achieve mutual natural human-computer interface, has great research value and application value. At present, the speech emotion recognition research focuses on speech signal preprocessing, speech feature extraction, dimensionality, speech emotion recognition algorithm.
Bias network is a statistical method to develop a decision-making method based on Bias proposed in 1988 by Pearl on the network. Bias with unique uncertain knowledge expression, rich probability expression ability, comprehensive incremental prior knowledge of the learning characteristics in recent years become the focus of theoretical research, is widely used in intelligence decision-making, pattern recognition, medical diagnosis and so on.
This paper selects the speech emotion recognition algorithm as the research focus, puts forward the basic combination of Bayesian network classifier is applied to speech emotion recognition. Based on Bayesian networks is studied, using Bayesian network to adjust the basic classification results, and proposed a Bayesian network generation algorithm based on confusion matrix, and the the improved algorithm is applied to the study on the speech emotion recognition, using support vector machine SVM and Naive Bayesian NB as the basic classifier, Bayesian network is used to adjust the results of classification in speech emotion database (Berlin, CASIA, SAVEE) to do the experiment, obtained better than the original algorithm classification accuracy.

【学位授予单位】：华南理工大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TN912.34

【参考文献】