Systems and Methods for Detecting Fraud in Spoken Tests Using Voice Biometrics
- Author(s):
- Papierman, Paul; Patibanda, Srinivas; Rajappan, Venu Valamparampil; Ferrara, Mary Ellen; Cline, Frederick; Wilcox, Michael T.
- Patent Issued:
- Oct 18, 2016
- Patent Number:
- 9,472,195
- Source:
- ETS Patent
- Document Type:
- Patent
- Family ID:
- 54191282
- Subject/Key Words:
- Patent, Active Patent, Spoken Language Assessment, Fraud, Biometric Voice Identification, Automatic Speech Recognition, Automated Scoring of Speech, Automated Scoring and Natural Language Processing, Spoken Language Assessment
Abstract
Systems and methods described herein automate imposture detection in, e.g., test settings based on voice samples. Based on user instructions, a processing system may determine at least one set of appointments, each having voice samples and a voice print, and a comparison plan for comparing the appointments. The comparison plan defines a plurality of appointment pairs. For each appointment pair, the system compares the associated first and second appointments by, e.g., comparing the first appointment's voice samples to the second appointment's voice print and generating corresponding raw scores, which may be used to compute a composite score. If the composite score satisfies a predetermined threshold condition for fraud, the system may determine whether flagging/holding criteria are satisfied by the raw scores. If the criteria are satisfied, a flag or hold notice may be associated with the appointment pair to trigger an appropriate system/human response (e.g., withholding the appointments' test results).