Deception detection in speech using bark band and perceptually significant energy features

This paper presents the initial results of analysis of nonlinear spectral features for classifying truthful and deceptive speech. These features are derived on a Bark scale based on the psychoacoustic masking property of human speech perception. Truthful and deceptive speeches are established a posteriori by a male speaker under jeopardy. Test results using significant energy features at Bark bands and a neural network have a potential to show delicate variations between truthful and deceptive utterances.