Deceptive Detection based on Audio Spectrum Analysis using Deep Learning

Lie detectors are intended to determine whether a person is telling the truth. Psychologists and computer scientists have done a lot of research to make this possible. The psychologist can determine if a normal person is telling the truth or not by his body language; but what if this person is a skilled actor who can make the task of a psychologist difficult. For that, the investigators are using lie detectors with vital signs such as heart rate, etc. This research proposes to design a system that detects lies through deep learning using sound features such as mel-spectrogram. The proposed model can detect deception with a substantial accuracy of 100% on the training dataset and 97 % on the testing dataset.