A Cloud-Powered Multimodal AI System for Deception Detection Using Facial, Vocal, and Textual Analysis

Traditional lie detection systems such as polygraph instruments rely heavily on physiological responses including heart rate, blood pressure, and skin conductance. These methods suffer from limitations including invasiveness, low generalizabil- ity, and high susceptibility to anxiety-induced distortions. With the rise of cloud computing and deep learning, multimodal decep- tion detection—combining facial micro-expressions, vocal stress cues, and linguistic inconsistencies—has emerged as a highly promising alternative. This paper presents VeriCloud, a scalable cloud-powered AI framework that integrates three independent modalities: facial analysis via landmark-based micro-expression extraction, speech analysis using MFCC-driven BiLSTM mod- els, and textual analysis using TF-IDF and logistic regression. Each modality is deployed as a microservice, enabling parallel processing and low-latency inference. A weighted late-fusion strategy aggregates the individual predictions to compute the final deception likelihood. Experiments show that multimodal fusion outperforms unimodal systems, achieving a combined accuracy of 92.3%, thereby demonstrating the superiority of cloud-integrated behavioral AI over traditional polygraph systems.