ID R&D awarded patent for text-dependent voice biometrics improvement

ID R&D has been granted a patent for text-dependent speaker recognition involving improved speed and accuracy of voice biometrics, even with small amounts of training data.

The patent for a ‘System for text-dependent speaker recognition and method thereof’ was issued by the U.S. Patent and Trademark Office (USPTO) on an application filed in 2019. It describes the use of a “high dimensional speaker model” generated from a voice sample to extract “a low dimensional p-vector,” which is then compared to the p-vector obtained from the speaker during the enrollment process. ID R&D says it shows the company’s passion for advancing biometrics performance and usability.

Most recently-developed voice biometrics systems are based on i-vector, ID R&D says in the filing, but they perform poorly with short utterances. Instead, the company proposes a Gaussian Mixture Model (GMM) speaker model estimated with Maximum a Posteriori (MAP) adaptation.

The patent comes on the heels of an announcement from ID R&D that a third-party test of its voice recognition technology shows a major breakthrough in biometric accuracy.

“We are an R&D-driven company with a commitment to delivering products that are either first-in-class or entirely unique,” said Alexey Khitrov, president of ID R&D. “We get excited about making things that were previously impossible possible. Voice biometrics have the ability to re-invent age-old processes and add tremendous value to emerging technologies from smart home devices to connected cars.”

The company has already filed several additional patent applications to extend the voice biometric system described in the just-granted patent.

Article Topics

biometrics  |  ID R&D  |  patents  |  research and development  |  speaker recognition  |  voice biometrics