Phase II year
2018
(last award dollars: 2021)
Phase II Amount
$3,248,807
Voice phishing ("vishing") is an increasingly serious threat that cost victims over $46 billion in 2016. Attackers pose as trusted callers using impersonation, voice mimicry, speech synthesis, voice conversion, Caller ID spoofing, and other techniques. Once victims trust the caller, the perpetrator can engage in fraudulent and criminal activities leading to financial losses and security breaches. The proposed research will result in a Real-time Enhanced Voice Authentication (REVA) system that detects many common forms of vishing, is available as a SaaS solution and mobile app, and serves to verify known callers in real-time. REVA builds on and extends techniques used in automated interactive voice response systems and audio forensics to detect signal alterations and to verify known speakers. REVA will monitor voice streams in real time and use a machine learning algorithm to continually estimate the likelihood that vishing is taking place, alerting users to dangers before they reveal sensitive information. This capability will be of great value to the security and defense communities, call centers, businesses, and consumers. The proposed architecture and machine learning approach give REVA the ability to learn to detect new types of attacks and to be updated as network characteristics and communication technologies evolve.
