Deep Neural Networks for Lie Detection with Attention on Bio-signals

After years of research, scientists have consistently proven that the ability of people to detect deception is no greater than chance. Accurate deception detection is crucial for law enforcement officials. The conventional polygraph test is far from perfect. It can be deceived by a trained individual quite easily. Micro-expressions are difficult to control or fake. They occur as fast as 1/15th of a second. Speech is another feature that can be used to catch a liar. They often speak in high or low-pitched voices due to tightening of their vocal cords, or pause to think before answering. In this paper, we have come up with a robust solution. It is a deep learning model that combines micro-expressions along with audio features and is used to detect lies. This proposed model is essentially a lie detector and has vast applications. For instance, it can assist law enforcement officials during interrogations and human resource personnel while conducting corporate interviews.