VigilAI Documentation
Everything you need to detect deepfakes and AI-cloned voices with VigilAI, by Abhishek Kumar.
Launch the free detector →Getting Started
VigilAI detects AI-generated voices, deepfake video and cloned audio. It runs in any modern browser, on mobile or desktop.
- Open ai.vigilai.online.
- Choose a method: live microphone, file upload or webcam liveness check.
- Wait a few seconds for the verdict and confidence score.
For best results use Chrome or Edge, allow microphone and camera permissions, and use clear audio with little background noise.
Scan Methods
| Method | Input | Best for |
|---|---|---|
| Live microphone | Mic recording | Suspicious phone or video calls |
| File upload | WAV, MP3, MP4, MOV | Voice notes, recorded calls, viral clips |
| Webcam liveness | Live camera | Checking a face in real time |
Tip: 10 to 30 seconds of clean speech gives more reliable results than a very short clip.
How Detection Works
VigilAI combines several independent signals into one weighted verdict, so a single weak signal does not decide the result.
Audio
Spectral features such as 40 MFCC coefficients (via LibROSA) are measured for variance. Real voices vary naturally; many synthetic voices are unusually uniform. Waveform and spectrogram views are available in the app.
Video
MediaPipe tracks 468 facial landmarks for blink rate (eye aspect ratio) and micro-jitter on the nose bridge. A 2D Fourier transform checks for missing high-frequency pixel detail typical of generated faces.
Combined verdict
Weighted signals produce a score. Higher scores mean more evidence of synthetic media.
Reading Your Results
- Likely real: signals look natural. Still verify high-risk requests.
- Suspicious: mixed signals. Do not act on money or data requests; verify another way.
- Likely synthetic: strong evidence of AI generation. Stop engaging and report it.
Compression, noise and low-quality recordings can affect scores. Treat results as a strong hint, not courtroom proof.
Privacy and Data
Do not upload content you are not allowed to share. Avoid sharing sensitive documents in scans. Optional sign-in can save scan history and downloadable reports to your account. For questions about data handling, email abhishek@vigilai.online.
If You Have Been Scammed
- Stop contact and do not send more money.
- Call your bank immediately to block or reverse the transfer.
- In India, report at cybercrime.gov.in or call helpline 1930 as soon as possible.
- Save call logs, screenshots and recordings as evidence.
Troubleshooting
- Mic or camera not working: allow permissions in the browser address bar and close other apps using them.
- Upload fails: check the format (WAV, MP3, MP4, MOV) and try a shorter file.
- Page does not load: hard refresh, try Chrome, or disable extensions.
Still stuck? Email abhishek@vigilai.online.
API for Developers
A REST API and SDK for fintechs, banks and platforms (KYC deepfake checks, voice verification) is under development. Early access and partnerships: abhishek@vigilai.online.
FAQ
Is VigilAI free?
Yes. The detector is free to use at ai.vigilai.online.
Which file formats are supported?
Audio: WAV and MP3. Video: MP4 and MOV. You can also record from your microphone or use the webcam liveness check.
How accurate is the result?
VigilAI gives a confidence score based on weighted signals. It is a decision aid, not legal proof. Always verify important requests through a second channel.
Does it work in Hindi and other Indian languages?
Yes. Audio analysis looks at acoustic patterns, not words, so it is language independent.
Can I use VigilAI during a WhatsApp or Zoom call?
Yes. Record the call audio or use the live microphone and webcam checks while on the call.
Is there an API for banks and fintechs?
A developer API for KYC and onboarding checks is being explored. Email abhishek@vigilai.online for early access.