Next-generation voice anti-fraud systems from microtremor theory to practical deployment in MVNO network

The monograph is devoted to developing a system for detecting telephone fraud, aggressive, and collection calls for MVNO networks. Amidst the growth of voice attacks using TTS and deepfake, traditional protection methods (STIR/SHAKEN, blacklists, voice biometrics) prove ineffective.

A novel approach is proposed based on analysis of Talk-Time Ratio and vocal fold microtremor (8–12 Hz) with its harmonics (16–48 Hz)—physiological stress markers present in live fraudsters and aggressive interlocutors but absent in synthesized speech. A B2BUA architecture with kernel (XDP/eBPF) and user space logic split is developed. An inflection point heuristic filters out 90–95% of legitimate calls before FFT analysis.

Experimental validation demonstrated 92% accuracy with 5–6 second latency. The system is cost-effective for MVNOs (2-year payback). Mechanisms for family control and Federal Law 230-FZ compliance are developed.