Deepfake Forensics

Flagging manipulated video and cloned voices for police evidence.

Source on GitHub (opens in a new tab)

The problem

Deepfake video and cloned voices are now realistic enough to mislead an investigation. Police need to know whether a clip has been manipulated, how confident that call is, and a record that will stand up as evidence.

How it works

I fine-tuned a pretrained video model on the provided dataset to classify frames as real or manipulated, and integrated a pretrained audio model to catch synthetic and voice-cloned speech, so the tool checks both what is seen and what is heard.

The system returns a verdict with a confidence score and heat maps over the manipulated regions of a frame, then generates a PDF report with metadata and cryptographic hashes. A private blockchain ledger keeps the chain of custody. It is split into a Python AI service, an Express API and a Next.js interface.

Highlights

  • Video and audio detection in one tool
  • Confidence scores and heat maps per frame
  • Tamper-evident PDF reports with hashes
  • Blockchain audit trail for chain of custody