Acoustic Indicators of Deception in Mandarin Daily Conversations Recorded from an Interactive Game

Abstract

Being able to distinguish the differences between deceptive and truthful statements in a dialogue is an important skill in daily life. Extensive studies on the acoustic features of deceptive English speech have been reported, but such research in Mandarin is relatively scarce. We constructed a Mandarin deception database of daily dialogues from native speakers in Taiwan. College students were recruited to participate in a game in which they were encouraged to lie and convince their opponents of experiences that they did not have. After data collection, acoustic-prosodic features were extracted. The statistics of these features were calculated so that the differences between truthful and deceptive sentences, both as they were intended and perceived, can be compared. Results indicate that different people tend to use different acoustic features when telling a lie; the participants could be put into 10 categories in a dendrogram, with an exception of 31 people from whom no acoustic indicators for deception were found. Without considering interpersonal differences, our best classifier reached an F1 score of 53.37% in distinguishing deceptive and truthful segmentation units. We hope to present this new database as a corpus for future studies on deception in Mandarin conversations.

Automatic Deception Detection using Multiple Speech and Language Communicative Descriptors in Dialogs

“…We evaluate the proposed framework on the Daily Deceptive Dialogues corpus of Mandarin (DDDM) [27] collected at the National Tsing Hua University, Taiwan. It contains 27.2 h of audio recordings of dyadic interaction of Mandarin native speakers with 96 unique speakers (gender is balanced).…”