Layer 2 — Cryptographic Voice Identity
Voice Verification
Speak the live challenge. Replays and clones cannot.
Sesim Cryptographic Voice Identity (Layer 2) verifies a customer with a dynamic spoken challenge — a fresh phrase from a 10,000+ random pool that defeats replay attacks. The voiceprint match plus ASR plus anti-spoof + liveness lands a single decision in under 500 ms; ZKVP and NIST PQC primitives integrated during pilot.
Private pilot
Live verification is restricted to active pilot customers. To request access for a caller-auth, mobile voice-login, or ATM voice-PIN pilot, contact us.
Request pilot access →How verification works
- Bank requests a challenge: dynamic random phrase, 60-second TTL, single-use nonce.
- Customer reads the challenge aloud (2–4 seconds).
- ECAPA-TDNN compares against the stored voiceprint hash; faster-whisper checks the spoken phrase matches the challenge.
- Anti-spoof + liveness models reject TTS clones, voice conversion, replays and pre-recorded clips.
- Single decision returned: allow / reject + reason codes + audit hash.
Why dynamic challenge matters
- Defeats replay: an attacker cannot reuse a recording — the phrase changes every time (claims 48 / 61).
- Defeats SIM-swap: the secret is the customer’s voice + the live phrase, not an SMS code.
- Coercion-aware: cross-signal from Layer 1 raises a silent alarm if the customer is under duress (claims 20 / 47).
KPI gates (pilot)
- FAR ≤ 0.1% on the pilot channel.
- FRR ≤ 2% on retry-aware funnel.
- Spoofing-rejection ≥ 95% across replay, TTS clone and voice conversion red-team sets.
- p95 verification latency ≤ 500 ms server-side.
Pilot at a glance
Duration
8–12 weeks
Fee
$25–40k creditable
Scope
One channel — caller-auth, mobile voice-login or ATM voice-PIN
Payment
50% kickoff + 50% on success-criteria sign-off