Resemble AI is raising the bar for inline in-call detection with new support for leading telephony codecs — G.711, G.729, AMR-WB, and Opus — combined with a significant accuracy breakthrough in detecting synthetic and manipulated speech across compressed audio streams.
Why it matters
Most voice fraud today rides over bandwidth-constrained codecs optimized for telephony. Detecting deepfakes or impersonation attacks in these conditions is notoriously challenging due to lossy compression, packetization, and bandwidth artifacts. Our latest release closes this gap, ensuring high-precision detection even on low-bitrate, real-world call traffic.
What’s new
- Codec-Aware Detection Models – Enhanced accuracy across narrowband and wideband codecs without requiring PCM-quality input.
- Improved Inline Reliability – Greater resilience to compression artifacts, jitter, and packet loss in SIP, SIPREC, and RTP streams.
- Enterprise & Carrier Scale – Optimized for high concurrency deployments in contact centers, SBC environments, and telco networks.
- Security Against Evasion – Models hardened against codec-level adversarial perturbations, closing common attack vectors.
The outcome
Enterprises and carriers can now deploy real-time, in-call detection with confidence that voice integrity checks remain accurate regardless of the codec in use. Whether protecting a financial services hotline, a carrier call path, or a government comms channel, detection remains reliable, fast, and inline.
About Resemble’s Deepfake Detection Model
Resemble AI’s Deepfake Detection Model is a multimodal system built to identify manipulated audio, images, and video with industry-leading accuracy. Trained on a wide range of generative model outputs, it can spot synthetic media even as new architectures and techniques emerge. The model operates in real time, making it ideal for securing customer interactions, financial transactions, and enterprise communications where trust is critical. By combining audio forensics, visual signal analysis, and cross-modal consistency checks, Resemble delivers a unified layer of protection against deepfakes, ensuring authenticity across every channel where fraud may appear.
