EDUWORKS CORPORATION — Department of Defense SBIR Phase II: SB163-006

Description

Eduworks has been developing the Real-time Enhanced Voice Authentication (REVA) mobile app and underlying algorithms to demonstrate real-time detection of telephonic voice phishing attacks. REVA’s capabilities are based on positive authentication of a caller’s voice, detection of voice conversion and speech synthesis algorithms (audio deepfakes), and model-drive analysis of the spoken content to flag suspicious activity. These capabilities aim to detect and prevent vishing attacks in real time before damage has been done. In this Sequential Phase II, Eduworks will explore novel semantic analysis capabilities that augment existing detection technologies with algorithms for attribution and characterization of calls. Attribution will link suspicious activity to a particular individual or group, and characterization will associate suspicious activity with intent and potential threat. These new capabilities enable more targeted warnings to users when a threat has been identified and facilitates investigative analysis by linking calls that can be attributed to misinformation campaigns or groups of callers disguising their voices. These novel capabilities will be further developed into two prototypes. Prototype 1 will demonstrate real-time semantic call analysis in which REVA monitors telephone call using new detection capabilities augmented during the proposed effort. REVA will perform semantic analysis of intent based on the spoken content to determine potential threats and issue targeted warnings to the user. As a result, warnings will be more relevant to the user and will avoid false positives, such as benign calls from appointment reminder services using synthetic speech. Attribution capabilities will also enable issuing warnings when the caller’s voice is associated with previously identified suspicious calls. Prototype 2 will demonstrate investigative analysis, search and triage. REVA will analyze, in non-real time, large numbers of recordings of voice audio streams. Using its semantic analysis capabilities, REVA will help triage investigative analyses of calls to reduce the volume of collected data to exploit and to prioritize suspicious calls warranting review by human analysts. REVA will facilitate attribution by linking calls where suspicious activity was detected to a particular individual or group, for example, groups of people communicating using voice disguise mechanisms. REVA will facilitate characterization of intent using knowledge-based analysis of the spoken content to determine potential threat. REVA, via semantic profiling, can further serve as a powerful search filter for analysts, extracting recorded calls that fit a given attack vector or tampering signature.