Lehigh University - CSE:Hirschberg

Julia Hirschberg

"Detecting Deception from Speech: Acoustic/Prosodic and Lexical Cues"

Thursday, March 1, 4:00 PM

Packard Lab - Room 466

Deception detection has enjoyed increasing interest in recent years. However, most current studies focus on visual and biometric cues. We describe studies of the acoustic, prosodic and lexical features of deceptive speech to see how speakers change both what they say and how they say it when engaging in deception. I will describe a large corpus of deceptive and non-deceptive speech collected in collaboration with SRI/ICSI and the University of Colorado at Boulder and describe machine learning experiments in which we attempt to predict deception from a wide range of lexical and acoustic/prosodic cues. I will discuss classification accuracy rates in the context of the performance of human judges on this data, examine the role personality factors in deception detection, some experiments with speaker-dependent models of deception, and future research.

Dr. Hirschberg is a professor in the Department of Computer Science at Columbia University. She received her PhD in Computer Science from the University of Pennsylvania, after previously doing teaching (at Smith College) and research in sixteenth-century Mexican social history (PhD from the University of Michigan). She worked at Bell Laboratories and AT&T Laboratories in Research from 1985-2003 as a Member of Technical Staff and a Department Head. She's editor-in- chief of Speech Communication and was formerly editor-in-chief of Computational Linguistics, the journal of the Association for Computational Linguistics (ACL), from 1993-2003; she was also on the ACL Executive Board. She's a fellow of the American Association for Artificial Intelligence. She serves on the Permanent Council for the Organisation of International Conferences on Spoken Language Processing (ICSLP) and she's president of the International Speech Communication Association (ISCA).

		Julia Hirschberg "Detecting Deception from Speech: Acoustic/Prosodic and Lexical Cues" Thursday, March 1, 4:00 PM Packard Lab - Room 466 Deception detection has enjoyed increasing interest in recent years. However, most current studies focus on visual and biometric cues. We describe studies of the acoustic, prosodic and lexical features of deceptive speech to see how speakers change both what they say and how they say it when engaging in deception. I will describe a large corpus of deceptive and non-deceptive speech collected in collaboration with SRI/ICSI and the University of Colorado at Boulder and describe machine learning experiments in which we attempt to predict deception from a wide range of lexical and acoustic/prosodic cues. I will discuss classification accuracy rates in the context of the performance of human judges on this data, examine the role personality factors in deception detection, some experiments with speaker-dependent models of deception, and future research. Dr. Hirschberg is a professor in the Department of Computer Science at Columbia University. She received her PhD in Computer Science from the University of Pennsylvania, after previously doing teaching (at Smith College) and research in sixteenth-century Mexican social history (PhD from the University of Michigan). She worked at Bell Laboratories and AT&T Laboratories in Research from 1985-2003 as a Member of Technical Staff and a Department Head. She's editor-in- chief of Speech Communication and was formerly editor-in-chief of Computational Linguistics, the journal of the Association for Computational Linguistics (ACL), from 1993-2003; she was also on the ACL Executive Board. She's a fellow of the American Association for Artificial Intelligence. She serves on the Permanent Council for the Organisation of International Conferences on Spoken Language Processing (ICSLP) and she's president of the International Speech Communication Association (ISCA).