ID R&D announces high accuracy in NIST voice biometrics testing as survey shows deepfakes misunderstood
ID R&D has announced the accuracy results of its voice biometric technology in the 2019 Speaker Recognition Evaluation (SRE): Conversation Telephone Speech (CTS) Challenge from NIST.
The 2019 SRE CTS Evaluation is the latest in a series of evaluations to provide a common test bed for measuring developments and supporting research into voice technology. There were entries from 51 commercial and academic research teams in the evaluation.
“Using our advanced modified x-vector algorithms, ID R&D was able to complete the 2019 NIST challenge with outstanding results and low Equal Error Rate (EER) on challenging telephone audio data in the Tunisian Arabic language,” comments ID R&D President Alexey Khitrov. “ID R&D is committed to building voice biometric technology with breakthrough performance in accuracy, speed, and footprint.”
ID R&D was also invited to share its methodology with the NIST community at the recent SRE workshop in Singapore, according to the announcement.
The company also recently announced a first-place finish detecting synthetic speech in the 2019 ASVspoof Challenge, which is more difficult task than many Americans seem to think. A survey from ID R&D shows that 36 percent of people in the U.S. are confident they can detect a computer-generated voice which is presented as a real human.
Just 30 percent are not confident they could tell the difference, by contrast, and 66 percent express concern that voice biometrics could allow someone to access their accounts by impersonating them.
The company notes that at the 2019 Network and Distributed System Security Symposium (NDSS), researchers from the University of California Riverside, University of Alabama at Birmingham, and Syracuse University shared evidence that how the brain processes authentic human speech and morphed speech may not be significantly different, possibly meaning the differences are indistinguishable at the neural level.
Asked how they would prefer to long into devices and accounts, 27 percent said they would prefer to use voice than an alphanumeric passcode, and 40 percent said they would prefer a highly secure but easy to use method such as voice or face biometrics to access sensitive accounts. A third said they would use a home voice assistant like Alexa to access account information if they were confident in the security of the transaction.
The survey also shows 27 percent would engage chatbots more if they were assured of security, and 32 percent would use chatbots more if they knew they could complete a transaction without being transferred to a customer service representative.
“This research shows that the biometric industry has a lot of work to do to educate consumers around legitimate security issues in voice technology,” says Khitrov. “Juniper Research estimates that there will be 8 billion digital voice assistants in use by 2023, yet our research suggests that despite such strong adoption, concerns about security are curbing what could be even stronger growth. Those of us in the biometric industry have a responsibility to educate consumers about the risks of deepfakes and synthetic voice, but also a real opportunity to educate consumers about the many benefits of biometrics, including improved security. Just as the consumers in the early 1990s were suspicious about online commerce but now can’t imagine life without it, we believe that once users learn how biometrics can better protect their data and accounts while delivering an all-around better experience across all applications, voice technology will see exponential growth.”
ID R&D will demonstrate its biometric technology, which it says is designed to prevent spoof attacks involving deepfakes and recordings, at CES in January.