Synthetic Voice Fraud in the Contact Center

Voice cloning has crossed the threshold from a niche technical capability to a commoditized fraud tool. Deepfake-as-a-Service platforms offer synthetic voice generation for under $10 per month, and modern voice clones achieve conversational latency low enough to be indistinguishable from a live caller in real time. Agents cannot hear the difference.

The threat extends beyond voice biometric systems. Contact centers that rely on knowledge-based authentication are equally exposed. Synthetic voices paired with fraud dossiers assembled from breached data can impersonate account holders convincingly enough to defeat KBA at scale, without any biometric system in the path at all. The attack vectors include:

• Account takeover via voice biometrics: synthetic voices impersonating account holders to pass active voice authentication
• KBA fraud: AI-generated voices combined with breached PII to answer security questions and gain agent trust
• Social engineering: synthetic voices used to manipulate agents into bypassing security procedures, disclosing account details, or authorizing transfers
• Replay attacks: recorded audio replayed to defeat active voice authentication

Knowledge-based authentication cannot distinguish a synthetic voice from a human one. Voice biometric matching without anti-spoofing layers shares the same blind spot, and that gap is what synthetic voice fraud is specifically designed to exploit.

Daon provides real-time synthetic voice detection that operates from the first seconds of a call, regardless of language, dialect, or the voice cloning model in use.

Detection works at the signal level. This means analyzing frequencies, phase relationships, and digital artifacts characteristic of AI-generated audio, rather than comparing against a voiceprint or enrolled identity. No enrollment is required, the capability is invisible to legitimate callers, and it operates across any language or dialect with no configuration changes. Known voice cloning models are detected with 99% accuracy, and the system adapts through machine learning as synthesis techniques evolve.

When deployed alongside voice biometric authentication, the defense is layered: voice biometrics are equipped with their own anti-spoofing stack, replay detection, synthetic speech detection, pre-seen audio detection, and distorted voice detection, with xDeTECH adding a dedicated signal-level detection layer on top. Voice biometric authentication integrates natively with Genesys, Cisco, Amazon Connect, Avaya, Five9, and NICE, deploying into existing contact center infrastructure without replacing it.