Voice cloning uses AI to replicate a specific person’s voice from a short audio sample, then generate new speech in that voice. Open-source models can produce convincing output from ten seconds or less of reference audio, and most ship without watermarks.

• A neural model is trained or conditioned on a target voice sample
• The cloned voice can speak any text with control over emotion, pacing, and emphasis
• Reference audio is sourced from podcasts, earnings calls, conference recordings, or social video

• Wire transfer fraud where a finance team member receives a "CEO" call authorizing an urgent payment
• Vendor impersonation calls to AP teams redirecting invoices
• Live video meetings paired with face swap for executive impersonation
• Vishing campaigns at scale using cloned executive or celebrity voices
• Bypass attempts against voice biometric verification at onboarding or in the contact center