학술논문
基于关联规则的语音情感中韵律特征抽取算法研究 / RESEARCH ON PROSODIC FEATURE EXTRACTION ALGORITHM IN SPEECH EMOTION BASED ON ASSOCIATION RULES
Document Type
Academic Journal
Author
Source
计算机应用与软件 / Computer Applications and Software. (9):42-77
Subject
Language
Chinese
ISSN
1000-386X
Abstract
近年来，情感计算一直是学术界研究的热点问题。语音情感识别作为情感计算的重要研究且涉及到人工智能、模式识别、机器学习等多个领域。针对语音情感识别中特征挖掘的复杂性，利用关联规则挖掘算法对语音特征中的韵律特征与所包含情感之间的关联关系进行研究。主要进行如下工作：（1）针对语音情感的特点，给出了情感频繁项集的概念；（2）提出基于关联规则的语音情感中韵律特征抽取算法（PFEA＿AR）；（3）在汉语情感数据集上进行相关实验，取得了85％的识别率，比fisher准则判别法的精度提高了10％。实验结果表明，通过关联规则算法所抽取的特征在降低维度的同时还能够有效提高情感分类精度，从而验证了新算法所抽取特征的有效性。
During recent years,affective computing has been the hot research topic in academia.Speech emotion recognition is the impor-tant study in affective computing,and is relevant to a couple of fields including artificial intelligence,pattern recognition,and machine learn-ing,etc.In light of the complexity of feature mining in speech emotion recognition,by using association rule mining algorithm we studied the pertinence relationship between the prosodic feature and the involved affect in speech feature,the main works done are as follows:(1 )for the features of speech emotion,we presented the concept of sentimental frequency item;(2)we proposed the association rule-based prosodic fea-ture extraction algorithm (PFEA_AR)in speech emotion;(3 )we carried out the correlated experiment in Chinese sentiment dataset,and achieved the recognition rate of 85%,1 0% higher than the discriminant using fisher rule.Result of experiment suggested that the features extracted by association rule algorithm can reduce the dimensionality while effectively improve the accuracy of sentiment classification,thus verified the effectiveness of the feature extracted by the new algorithm.
During recent years,affective computing has been the hot research topic in academia.Speech emotion recognition is the impor-tant study in affective computing,and is relevant to a couple of fields including artificial intelligence,pattern recognition,and machine learn-ing,etc.In light of the complexity of feature mining in speech emotion recognition,by using association rule mining algorithm we studied the pertinence relationship between the prosodic feature and the involved affect in speech feature,the main works done are as follows:(1 )for the features of speech emotion,we presented the concept of sentimental frequency item;(2)we proposed the association rule-based prosodic fea-ture extraction algorithm (PFEA_AR)in speech emotion;(3 )we carried out the correlated experiment in Chinese sentiment dataset,and achieved the recognition rate of 85%,1 0% higher than the discriminant using fisher rule.Result of experiment suggested that the features extracted by association rule algorithm can reduce the dimensionality while effectively improve the accuracy of sentiment classification,thus verified the effectiveness of the feature extracted by the new algorithm.