File size: 1,884 Bytes
18906db e597742 18906db 07dbf1f 18906db e597742 18906db e597742 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 | ---
title: DEEP FAKE DETECTOR
emoji: 🎭
colorFrom: blue
colorTo: purple
sdk: gradio
sdk_version: 6.14.0
app_file: app.py
pinned: true
license: mit
short_description: Multimodal deepfake detection across 5 modalities
---
# Multimodal Deepfake Detection System
A modular AI system for detecting fake content across 5 modalities — text, audio, image, video, and lip-sync — using hybrid CNN-Transformer-LLM architectures.
## Live Demo
Try the live system using the tabs above:
- **Text Verification** — Detects fake news using SERP API + DeBERTa + Llama-3.3
- **Audio Forensics** — Identifies AI-cloned voices using Wav2Vec2 + AST + CLAP
- **Image Analysis** — Detects deepfake images using ResNet-18 + Vision Transformer
- **Video Forensics** — Analyzes video deepfakes with multi-frame voting + OpenCLIP
- **Multi-Modal Lip-Sync** — Detects audio-video mismatch using LSE-Net + Claude reasoning
## Performance
| Modality | Accuracy |
|----------|----------|
| Text | 99% |
| Image | 92% |
| Audio | 91% |
| Video | 91% |
| Lip-Sync | 89% |
## Tech Stack
- **Frameworks:** PyTorch, Hugging Face Transformers, Gradio
- **Vision Models:** Vision Transformer (ViT), ResNet-18, OpenCLIP
- **Language Models:** DeBERTa-v3, Llama-3.3, Claude 3.5 Sonnet
- **Audio Models:** Wav2Vec2, AST, CLAP, Whisper
- **Datasets:** FaceForensics++, FakeAVCeleb, ASVspoof
## About
This system was developed as part of MS thesis research at GIFT University, Pakistan.
**Author:** Aneela Pervez
**Institution:** GIFT University, Department of Computer Science
**Supervisor:** Dr. Muhammad Ziad Nayyer
## Links
- **GitHub Repository:** https://github.com/ANEELA-PERVEZ
- **LinkedIn:** https://linkedin.com/in/aneela-pervez-95472b310
- **Paper:** Coming soon on arXiv
## Citation
If you use this system in your research, please cite:
``` |