期刊名称：International Journal of Innovative Research in Science, Engineering and Technology
印刷版ISSN：2347-6710
电子版ISSN：2319-8753
出版年度：2016
卷号：5
期号：3
页码：3245
DOI：10.15680/IJIRSET.2016.0503100
出版社：S&S Publications
摘要：The main goal of this work is to show how a set of several time-domain features, as extracted fromspeech, can contribute for the Emotion Recognition in Speechproblem (ERS). Moreover, their fluctuations among acollection of spoken voiced events, which form multiple spoken utterances baring five distinct emotions are studied anddisplayed for analysis. In more details, the study compares the rate of change in time and uses their speakerindependent inter-sentence derivatives while comparing its performance to the classical featureapproach. Differentclassification methods were employed such as Naive Bayes, Support Vectors and k-NN. In the spoken database therewere clearly five distinctive emotional classes namely: happy (H), angry (A), sad (S), neutral (N) and fear (F), and forshort this modelis referred to as HASNF. The system shows a very successful recognition rate of 91.67% when usingthe chosen time-domain feature vector. In contrast, the rate of change of these parameters is not as promising, and canbe concluded that change of attributes is not as emotion-dependent as their parameters. The performance of the secondsystem topped 44.17% correct recognition rate, which is not successful for the five-classtime-domain feature caseconsidered here (HASNF).
关键词：Emotion; Speech; Recognition; Emotion recognition in speech; Time-domain features from speech.
