ReLiDSS: Novel lie detection system from speech signal

Abstract

Lying is among the most common wrong human acts that merits spending time thinking about it. The lie detection is until now posing a problem in recent research which aims to develop a non-contact application in order to estimate physiological changes. In this paper, we have proposed a preliminary investigation on which relevant acoustic parameter can be useful to classify lie or truth from speech signal. Our proposed system in is based on the Mel Frequency Cepstral Coefficient (MFCC) commonly used in automatic speech processing on our own constructed database ReLiDDB (ReGIM-Lab Lie Detection DataBase) for both cases lie detection and person voice recognition. We have performed on this database the Support Vector Machines (SVM) classifier using Linear kernel and we have obtained an accuracy of Lie and Truth detection of speech audio respectively 88.23% and 84.52%.

Cite this work

No public copy is stored for this record, so only the metadata is cited.

Stored only when both papers are in the corpus — 48 citations exist upstream.

Papers that cite the same references — the relational fallback while embeddings are still backfilling.