LDS–LVAT: Lie Detection System–Layered Voice Technology
K. M. Veena, Kanak Meena, D. Rajalakshmi, M. Fathima, K. Selvi · 2024
When dealing with criminal cases, investigators find detecting lies and dishonesty to be a considerable issue. In comparison to normal human conduct, the process of identifying a liar has a higher proportion of importance in terms of external behavior and cognitive power of the brain. Mel-frequency cepstrum coefficients (MFCC) approach extracts distinctive features from the original electroencephalogram data and utilizes them in conjunction with the neural network (NN) methodology for training and evaluation. Existing lie detection systems rely on physiological and behavioral factors, resulting in limited effectiveness. However, the pursuit of a computational model for automating lie detection has not been extensively explored. Researchers have recently focused on training machine learning models, including sequential NNs, solely using acoustic data from speech to enhance the accuracy of lie detection. The MFCC, energy envelopes, and pitch contours are constructed using a balanced data set of deceptive and non-deceptive speech recordings taken from a two-person deception game. This model’s highest accuracy for lie detection is 85.8%. The layered voice analysis is a new technique where an analysis of the same is done. It is capable of detecting and quantifying a wide range of psychological reactions that are suggestive of shifts: in the tested party’s perception, and it will warn the trained operator to follow its indications and leads.