Solution
Voice liveness verification with explainable decision trace
Sonotheia provides voice liveness verification by analyzing physics-based signal metrics on live and recorded audio streams. Every liveness verdict includes sensor-level measurements, triggered rules, and a documented threshold history, not just a pass/fail score.
Liveness checks that return only a boolean fail teams under audit. Sonotheia's governance layer preserves reproducibility so you can show exactly what was active when a voice-driven decision was made.
What is voice liveness verification?
Voice liveness verification confirms that audio originates from a live human speaker rather than a replay, synthesis, or conversion attack. Sonotheia applies multiple complementary sensors (spectral texture, envelope dynamics, and phase-coherence measurements) and fuses them with documented weights rather than a single opaque embedding.
Voice liveness detection for banks and credit unions
Banks and credit unions use voice liveness detection for banks when callbacks, wire authorizations, or credential resets still rely on voice influence. Sonotheia pairs liveness verdicts with decision trace and calibration history so BSA/AML and fraud ops can defend the control under examiner review.
What does explainable liveness mean in practice?
Each verdict ties to concrete measurements and documented thresholds. Supervisors see which sensors contributed, what rules fired, and why the outcome was reached. This supports FINRA vendor oversight, EU AI Act transparency duties, and internal model-risk reviews.
How is voice liveness deployed in production?
Deployments run on-prem or in customer-controlled private VPC containers. Audio never leaves the partner boundary for model training. Blind in-situ calibration is the default: threshold sweeps report aggregate non-sensitive statistics (EER, minDCF, ROC) without exporting customer audio.
Frequently asked questions
- Is this the same as passive voice biometrics?
- No. Sonotheia does not enroll or match voiceprints. Liveness verification analyzes the acoustic signal for spoof indicators without creating biometric templates.
- What is enterprise voice spoof detection?
- Enterprise voice spoof detection evaluates whether audio is synthetic, converted, or replayed before a high-trust action proceeds. Sonotheia returns explainable sensor outputs and governance artifacts so fraud and compliance teams can review the verdict, not only a boolean.
- Does voice liveness work on phone calls?
- Yes. Multi-codec calibration covers wideband, G.711, and AMR-NB telephony chains. Performance varies by codec; pilot calibration establishes channel-specific thresholds. Real-time operations apply within evaluated pilot contexts.
- What API outputs does liveness verification return?
- Structured JSON with verdict, confidence band, per-sensor contributions, decision trace, and evidence-contract hashes suitable for audit logs.
- Can liveness run on uploaded recordings?
- Yes. Upload and live-mic workflows are supported in the Shema product surface for pilot evaluation.
- How do you handle model drift?
- Baseline locking and versioned calibration history preserve reproducibility. Updates are logged so teams can show what was active at any point in time.
- What is Shema?
- Shema is Sonotheia's consumer-facing detection portal (live mic, upload, voice call modes) used for pilots and demonstrations at shema.sonotheia.ai.
Ready to evaluate Sonotheia for your voice channel?
Request a demo