Distinguishing Deceptive Speech from Truthful Speech using MFCC Features

扫码查看

原文链接

NETL

外文摘要：This paper presents the results from a preliminary study to distinguish deceptive speech from non-deceptive speech using mel frequency cepstrum coefficients (MFCCs)。 Cepstral coefficients for the utterances of 'No' from known deceptive speech and ground truth from a speaker under criminal investigation were used as features in a neural network。 Result from the network trained with two pairs of utterances indicated correct recognition rates ranging from better than 60 percent to close to 100 percent using delta cepstrum and difference cepstrum features; MFCCs， however， yielded accurate recognition in just 45 percent to 61 percent of the test cases。

外文关键词：

Feature ExtractionCritical bandMel-frequency WarpingCepstrumDeceptive SpeechArtificial Neural Network

作者：

MUHAMMAD SANAULLAH、KALIAPPAN GOP ALAN

展开 >

作者单位：

Electrical and Computer Engineering Department Purdue University Calumet Hammond, IN 46323 USA

会议名称：

WSEAS international conference on applications of electrical engineering;WSEAS international conference on applications of computer engineering;International conference on communications and information technology;International conference on circuits, syst

会议地点：

Cambridge, MA(US)

会议母体文献：

Recent advances in electrical and computer engineering

页码：

167-171

出版时间：

2013