ID R&D and Synaptics partnership to bring voice biometrics to smart home edge devices
ID R&D and Synaptics have partnered to integrate voice biometrics with the VideoSmart VS600 platform to enable fast and accurate identification, authentication, and personalization on edge computing devices for smart home applications.
The biometric capability is integrated with the NPU of the VideoSmart VS600 System-on-Chip (SoC), which also uses SyNAP technology to enable the biometric algorithms to run on edge computing devices. The NPU supports trillions of operations per second to run machine learning models at the network edge. By tapping into the NPU’s capabilities, ID R&D says it can run the sophisticated algorithms that deliver high-accuracy voice biometrics, as well as liveness detection to detect spoof attacks, with an extremely small footprint.
The companies say the result is a first-of its kind capability, with 10 times the speed of traditional systems running the algorithm on the CPU and both CPU and RAM utilization are dramatically decreased. The combined solution works with home devices like set-top boxes, smart speakers, displays, and security systems.
“This is the first known commercial application of voice biometrics with liveness on an NPU for smart home applications. The speed of the voice biometric processing is faster than real-time — meaning we can identify a speaker in a family group even before they are finished speaking a wake word and a command to a home-based smart device,” says John Amein, SVP at ID R&D. “Previous systems have been limited by the capacity of the core CPU, which is not optimized for neural networks. This integration makes it possible to implement more sophisticated algorithms, opening up new use cases. We expect to enable voice for higher security levels necessary for use cases like payments, highly responsive identification of a person among a group for personalized responses, and a more natural user experience for authentication.”
The ID R&D software build is expected to reach availability on Synaptics’ VS600 development kids in February, 2021.
“The ability to instantly and reliably recognize a specific user on a shared device enables seamless personalization of controls and content without the frustrations associated with remembering and typing passwords,” comments Synaptics VP of Marketing for Edge Computing Vineet Ganju. “The solution also enables restricted access where needed, such as enabling parental controls.”
ID R&D also recently brought its voice biometric technology to WhatsApp for Busines through a new partnership.
The Intel x86 architecture that still dominates the desktop and datacenter markets were designed as general purpose “central processing unit,” or CPU. These designs have heat and power consumption characteristics that make them inefficient for the compute requirements of AI and ML models. This has helped drive a host of startups and incumbent chip vendors such as Chooch.AI and Deep Vision to design processors specifically for the data throughput requirements of AI. Synaptics’ neural network processing unit (NPU) on its VS600 chip provides offloading of AI models from the CPU; partnering with ID R&D provides a good example of the value of the NPU’s performance, showcasing specialized algorithms that offer increased accuracy for voice biometrics and anti-spoofing as an example.
Jim Davis, Principal Analyst, Edge Research Group