VoiceVantage™ VoiceCheck™ Verification Capability Analysis using AI Deep Fake Voice Applications

Voice-based biometric authentication systems such as VoiceVantage™ VoiceCheck™ may be increasingly vulnerable to AI-generated synthetic voice cloning, commonly known as deep fake technology. To evaluate the resilience of VoiceCheck™ against five popular deep fake applications, FakeYou, PlayHT, Descript, ResembleAI, and ElevenLabs, a series of controlled tests were carried out across two phases. Both internal and external attack scenarios were considered.
It was found that only deep fake attacks on VoiceCheck™ voiceprints created using 8kHz landline veriphrase recordings are minimally vulnerable to deep fake attacks, and only if VoiceCheck™ validation guidelines are not followed. All higher quality voiceprints remained secure within the VoiceCheck™ recommended configurations. The alternative of setting a somewhat higher pass thresholds as opposed to referring a low score to a Help Desk for validation was considered. It proved to be a viable alternative.