Introducing Telephony Optimized Deepfake Detection Model

Resemble AI has enhanced its in-call detection capabilities by supporting telephony codecs such as G.711, G.729, AMR-WB, and Opus, alongside achieving a breakthrough in accurately detecting synthetic and manipulated speech in compressed audio streams. These advancements are crucial in combating voice fraud, including deepfakes and impersonation attacks, which are challenging to detect due to lossy compression and bandwidth artifacts in telephony codecs. The new codec-aware detection models offer improved accuracy across narrowband and wideband codecs and provide greater resilience to compression artifacts, jitter, and packet loss in SIP, SIPREC, and RTP streams. Optimized for high concurrency deployments, the system ensures real-time detection in various environments, including contact centers and telco networks, by hardening the models against codec-level adversarial perturbations. Resemble AI’s Deepfake Detection Model, a multimodal system capable of identifying manipulated audio, images, and video, offers a comprehensive layer of protection against deepfakes by integrating audio forensics, visual signal analysis, and cross-modal consistency checks, ensuring reliability and authenticity in safeguarding customer interactions and communications.