基于SVM的语音情感识别研究  

Research on speech emotion recognition based on SVM

作　　者：胡洋[1] 吴黎慧[1] 高磊[1] 蒲南江[1] 

机构地区：[1]中北大学信息与通信工程学院,太原030051

出　　处：《电子测试》2011年第9期25-28,共4页Electronic Test

摘　　要：随着计算机技术的发展,人们对和谐人机交互的要求不断提高,这就要求计算机能理解说话人的情感信息,即能进行语音情感识别。本文提出了一种基于支持向量机(SVM)的语音情感识别方法,主要对人类的6种基本情感:高兴、惊奇、愤怒、悲伤、恐惧、平静进行研究。首先对自建语音情感数据库的情感语句提取特征,然后运用序列前向选择(SFS)算法选取最优特征集,最后通过二叉树支持向量机(BT-SVM)进行实验,取得了比较满意的识别率,证实了这种方法的可行性。With the development of the computer technology,people enhance unceasingly to the harmonious manmachine interaction request.This request the computer can understood the speaker＇s emotion information,namely can carry on the speech emotion recognition.In this paper,a Support Vector Machine （SVM） method on the speech emotion recognition is presented.6 kinds of basic emotions：happy,surprise,anger,sad,fear and clam is studied.First,the feature of the statement of self-build emotional speech database are extracted,then the optimal characteristic sets are selected by using Sequence Forward Selection （SFS） algorithm,finally we put forward the experimentation through the binary tree support vector machine and obtain the quite satisfactory recognition rate.This also confirms the feasibility of this method.

关 键 词：语音情感识别 支持向量机 特征提取 

分 类 号：TN912.34[电子电信—通信与信息系统]