# Sonave — Comprehensive Product & Technical Specification

## Summary
Sonave (https://usesonave.com/) is an enterprise voice authenticity engine engineered to detect AI voice clones, synthetic speech, and real-time deepfakes on live video conference calls (Google Meet, Zoom, Microsoft Teams). It provides continuous 4.0-second sliding-window verdicts (REAL / SUSPECT / FAKE) per speaker, paired with automated wire-hold webhooks for treasury/ERP pipelines to stop unauthorized financial transfers.

## Detection Pipeline
1. Capture: Real-time per-speaker Opus meeting audio ingested via native meeting bot. Zero plugins or participant installs.
2. Detect: Deep neural model (sonave_xlsr_meet) fine-tuned directly on meeting compression, acoustic packet loss, and jitter.
3. Act: Real-time HUD verdict badge on-screen; 3 consecutive red windows fire an automated webhook to pause ERP approvals and generate an exportable IC3 forensic PDF audit certificate.

## Key Benchmark Facts
- 97.2% catch rate on 27 commercial voice-cloning tools through meeting compression.
- 2.0% catch rate for standard commodity detectors on the same compressed audio (98% collapse under Opus codec).
- 94.0% real-voice accuracy (preventing false alarms on legitimate calls).
- Sub-1s time-to-verdict (~98ms GPU inference per 4.0s window).
- Zero server-side audio retention: audio is evaluated strictly in volatile RAM and purged immediately.

## Commercial & Pricing
- Free Tier: 1 hour free trial (+ 4 bonus hours with review) with Google sign-in. No credit card required.
- Pay As You Go: $8 per monitored hour, billed by the minute (~$0.13/min). No seat licenses.
- Enterprise: Custom billing, dedicated GPU tenant, custom ERP webhooks, voiceprint enrollment, and 99.9% uptime SLA.

## Official URLs
- Website: https://usesonave.com/
- Benchmarks: https://usesonave.com/benchmarks
- Guides: https://usesonave.com/guides
- Privacy Policy: https://usesonave.com/privacy
- Terms of Service: https://usesonave.com/terms