# Voting-based Multimodal Automatic Deception Detection

**Authors:** Lana Touma, Mohammad Al Horani, Manar Tailouni, Anas Dahabiah, Khloud Al Jallad  
**arXiv:** [2307.07516](https://arxiv.org/abs/2307.07516)  
**Date:** 2023-06-30  
**Summary page:** https://www.summarizepaper.com/en/paper/2307.07516v1/

## Key points

- Researchers use machine learning and deep learning techniques for Automatic Deception Detection
- The paper proposes a method for detecting deception from videos using audio, visual, and lexical features
- The proposed solution consists of three models: CNN for image detection, SVM on Mel spectrograms for audio detection, and Word2Vec on SVM for text detection
- Experiments were conducted on two datasets: Real-life trial dataset by Michigan University and Miami University deception detection dataset
- Results show that the proposed solution outperforms state-of-the-art methods:
- 97% accuracy for image detection
- 96% accuracy for audio detection
- 92% accuracy for text detection

## Summary

In the field of Automatic Deception Detection, researchers have been using machine learning and deep learning techniques to detect deception. In this paper titled "Voting-based Multimodal Automatic Deception Detection," the authors propose a method for automatically detecting deception from videos using audio, visual, and lexical features. The proposed solution consists of three models: a Convolutional Neural Network (CNN) for detecting deception from images, a Support Vector Machine (SVM) on Mel spectrograms for detecting deception from audio, and Word2Vec on SVM for detecting deception from manuscripts. The video samples are split into frames of images, audio, and transcripts. To evaluate the performance of their proposed solution, experiments were conducted on two datasets: the Real-life trial dataset by Michigan University and the Miami University deception detection dataset. The results show that their solution outperforms state-of-the-art methods. On the Real-Life Trial Dataset, the best results achieved were 97% accuracy for image detection, 96% accuracy for audio detection, and 92% accuracy for text detection.