The Problem
Real-time voice cloning has moved from research demo to commodity attack tooling. A convincing synthetic voice can now be injected into a live call to defeat voice authentication, authorize fraudulent transfers, and impersonate executives and family members — and passive acoustic detectors are locked in an escalating arms race against ever-more-realistic generators.
The Solution. CALLGUARD is software middleware that verifies a live caller is a real human on a real path — not a synthesis pipeline interposed in the call. It issues an active challenge and analyzes the timing of the response across signaling and media layers, isolating the processing latency an injected voice-conversion or text-to-speech stage necessarily adds. Because it reasons about call timing rather than audio artifacts alone, its detection power does not decay as generator realism improves.