Phychoacoustic Masking of Delta and Time -Difference Cepstrum Features for Deception Detection

摘要

This paper presents the test results of analyzing mel frequency cepstrum coefficient (MFCC), delta and difference cepstrum features to detect and distinguish the truthful and deceptive speech. The features are extracted based on the psychoacoustic masking property of human speech and how it is perceived. Truthful and deceptive speeches are preset based off a guilty male speaker in police custody. Delta cepstrum and time-difference cepstrum features at triangular critical bands filter and a neural network show the distinctions that determine whether an utterance is truthful or deceptive. In this paper, we analyze the extracted MFCC, delta cepstrum and time-difference cepstrum features to see how stress in speech accurately conveys human speech emotion and deception. Finally, we feed the data into an artificial neural network model to test out the results.