ID R&D update optimizes voice biometric solution for edge devices
A new version of the IDVoice biometric solution has been released by ID R&D, with what the company calls breakthrough optimization for edge devices, delivering high accuracy with an AI-based SDK under 1MB in size.
The small footprint of IDVoice version 2.11 enables voice biometrics implementation on chips and IoT devices at the edge for secure and personalized user experiences, according to the company announcement. The update features higher accuracy and speed with modified algorithms for the company’s unique modified x-vector approach, building on the technology which delivered strong results in NIST’s recent Speaker Recognition challenge. ID R&D’s biometric engine was previously 40MB, but can now run on smaller and less powerful devices with limited computing resources. IDVoice 2.11 is also available in special fusion versions, the company says, which can be optimized for specific use cases and hardware, from small edge devices to large call centers.
“With rising customer demand for conversational interfaces, ID R&D is committed to improving deployment, performance, and the user experience, no matter the channel or application,” states ID R&D Chief Science Officer, Konstantin Simonchik in the press release. “With IDVoice v2.11, developers can offer advanced voice biometric authentication on even the smallest device, opening up a whole new world of possibilities for securing the connected world.”
Mobile banking, call center fraud detection, and real-time passive call center authentication applications of IDVoice are currently in operation, and the software is built on ID R&D’s biometric engine architecture for channel, language, and age variance independence, with resilience to environmental noise. So far, most voice biometrics deployments are found in call centers and other relatively traditional compute environments.
IDVoice includes wrappers in C++, Python and Java, and ships as a Docker image to make adoption and integration easy for developers. It supports iOS, Android, Linux, and Windows, and is available for both text-dependent and text-independent applications. The company says the software also includes “diarization,” which detects and separates multiple voices for accurate authentication in audio streams with more than one person.
ID R&D’s passive facial biometric liveness technology recently received Level 1 Presentation Attack Detection (PAD) certification from iBeta.