Changelog

Updates and improvements to Resemble AI
Back to changelog
June 22, 2026

Audio DFD model update — 4% EER

Deployed a new audio deepfake detection model (W2V2 + Mamba + LoRA + LayerNorm). Equal error rate dropped from 7% to 4%, precision rose to 96% and recall to 94%, NVIDIA Magpie TTS detection reached 99%, and VoIP-with-packet-loss accuracy improved from 88% to 94%.