# Sonave

> Live deepfake-voice detection inside your meetings — every speaker gets a REAL / SUSPECT / FAKE verdict. Sonave watches the voices in your meeting and tells you, in real time, whether each one is human. Real-time per-speaker audio is streamed to a detection model trained on real meeting-codec audio; the side panel shows a live authenticity meter per speaker and a room-level verdict. When a voice scores in the red band for three consecutive windows, Sonave raises a wire-hold incident — with a webhook that can pause a payment approval and a one-click forensic report for compliance. Built for finance teams approving wires on calls, and for anyone who needs to know the voice on the other end is real. Free tier: 1 hour free (+ 4 bonus hours with review). Then $8 per monitored hour. Enterprise: usesonave.com.

Key facts (deployed model, benchmarked 2026-08-12; methodology at /benchmarks):
- 97.2% catch on 27 unseen commercial voice-clone tools through meeting audio (a commodity open-source detector catches 1.9% on the same clips)
- 94.0% real-voice accuracy through the Opus meeting codec
- 58.7% catch at 93.3% real-voice accuracy on In-the-Wild (the honest ceiling; verdicts are a second factor alongside callbacks, not a replacement)
- Sub-1s time-to-verdict powered by 98ms GPU inference
- Pricing: free 5 monitored hours/month, then $8 per monitored hour, self-serve

## Pages
- [Home](https://usesonave.com/): product, how it works, pricing, FAQ
- [Benchmarks](https://usesonave.com/benchmarks): full results + methodology
- [Guides](https://usesonave.com/guides): detecting deepfake voices on live calls, CEO voice-fraud anatomy, what detector accuracy numbers mean
- [Full LLM Documentation](https://usesonave.com/llms-full.txt): comprehensive technical and architectural specs
- [Privacy](https://usesonave.com/privacy) · [Terms](https://usesonave.com/terms)