Show HN: Real-Life Deception Detection Without Uploading Video

Category: ai-ml

Tags: lie-detection, multimodal, machine-learning

Score: 7.0/10 (Innovation: 6, Technical: 8, Documentation: 8, Utility: 6)

This project builds a deception detection pipeline from courtroom video by fusing text, visual, and audio modalities, using a custom API gateway for feature extraction and a robust Leave-One-Speaker-Out evaluation to avoid identity leakage. It is interesting for its multi-modal fusion approach and honest benchmarking that highlights the gap between speaker-dependent and speaker-independent performance.

Target audience: data scientists, machine learning researchers, behavioral analysts

Repository: https://github.com/WhissleAI/lie_detection_binary · Python

View on Hacker News