Speech Analysis using Modulation-Based Features for Detecting Deception

Abstract

This paper reports on the initial results of analysis of speech features for detecting deception from speech utterances. The features are derived from the amplitude- and frequency-modulation (AM and FM) behavior of speech centered at different nominal formant frequencies. Analysis of truthful and deceptive speech, both established a posteriori, by a male speaker under jeopardy shows that Teager energy-related features, amplitude envelope-based parameters, and formant variations and bandwidths at selected frequencies have a potential to show subtle variations indicative of deception. For comparison, trajectory of the fundamental frequency of voicing was found to show relatively minor variations between truthful and deceptive speech.