VigilAI detects AI-generated voices and deepfake video calls in real-time — before scammers can trick you.
// detection engine
Three independent signal layers cross-verify each other — a single suspicious result triggers the alert.
Extracts 40 MFCC coefficients from audio. AI voices have unnaturally low spectral variance — we catch that instantly.
Tracks 468 facial landmarks via MediaPipe. Deepfakes rarely blink naturally — our model checks eye aspect ratio across every frame.
Deepfake masks "shimmy" slightly. We track nose bridge movement variance and Fourier spectral energy — both giveaways of generation artifacts.
// capabilities
Every detection module targets the specific artifacts that AI generation leaves behind.
Record directly from your microphone during a suspicious call. Analyze in real-time as it happens.
Audio · MFCCUpload any .wav or .mp3 audio — or .mp4 / .mov video — for a deep-scan verdict with confidence score.
WAV · MP3 · MP4One-click live webcam scan. The AI monitors 468 facial landmarks across every frame.
MediaPipe · 468 ptsReal humans blink. Deepfakes don't — at least not naturally. We check eye aspect ratio frame by frame.
Biological SignalAI-generated faces are too smooth. We use 2D FFT to detect missing high-frequency pixel data.
Spectral · DFTThree detection layers cross-verify every signal. One red flag is enough. Results in under 5 seconds.
Local · Private// try it yourself
The full detection system runs on Streamlit. Click below to launch the app and scan any voice or video for AI manipulation — no install needed.
▶ Launch Detection App// get in touch
Questions, partnerships, or want to report a deepfake? Reach out directly.
// the builder