Remote exams and interviews are easy to cheat.
Someone else sits the exam or interview. The candidate reads a pre-written or ChatGPT answer aloud. A helper off-camera feeds lines they simply repeat. Webcams and honor codes don't catch any of this — VivaGuard does, by listening to who is speaking and how.
Two questions, asked of every recording.
Is this really the enrolled person — and are they genuinely answering?
Capture the real voice
The candidate reads a random sentence aloud within a short, timed window — a liveness step that captures a genuine, live reference voice.
Confirm identity
Any later recording is matched against the enrolled voiceprint. Same person or not — a clear accept/reject with a similarity score.
Detect reading
Scores speech naturalness — disfluencies, pauses, pacing, AI-text patterns — to tell genuine thinking-aloud from a read or fed answer.
One approach, every kind of cheat.
Built for institutions that can't send data away.
Identity + naturalness
Combines voiceprint matching with speech-naturalness analysis — catching stand-ins, read-aloud, and coached answers.
Private by design
No audio, transcript, or result ever leaves your server. Zero third-party calls, zero per-call cost.
Self-hosted
One-command Docker install. Runs CPU-only and fully air-gapped after setup.
Easy to integrate
A clean REST API plus a built-in web console — drop it into any exam workflow.
A clean REST API and a built-in console.
# Verify a recording against an enrolled candidate curl -X POST $BASE/api/verify \ -H "Authorization: Bearer $KEY" \ -F person_id=alice \ -F file=@answer.m4a # → { "verdict": "GENUINE", "similarity": 0.79, "match": true }
Bring integrity to every exam and interview.
See VivaGuard run on your own infrastructure — no data leaves your server.
Request a demo