June 6, 2016 -
Ambitious startup KnuEdge launched out of a ten-year stealth phase on Monday with a new neural computing technology and a set of noise-tolerant voice authentication products.
Led by founder and CEO Dan Goldin, who previously was NASA’s longest-serving head, KnuEdge has already raised $100 million in private funding and $20 million in revenue while in stealth mode, and says its KnuVerse voice authentication system opens up a range of real-world applications.
“We are not about incremental technology. Our mission is fundamental transformation,” said Goldin. “We were swinging for the fences from the very beginning, with intent to create next-generation technologies that will in essence alter how humans interact with machines, and enable next-generation computing capabilities ranging from machine learning to artificial intelligence.”
KnuVerse provides a set of algorithms for passive or active voice authentication in noisy real-world environments, which the company says unlock a range of practical, secure, and frictionless authentication uses. In addition to KnuVerse, KnuEdge offers neural computing acceleration technology with advanced machine learning and pattern recognition capabilities under the KnuPath brand.
KnuPath’s product is LambaFabric, which consists of new processor hardware and software, networked in a new style of architecture inspired by biology. First generation LambaFabric processors include 256 DSP cores and a router, and are connected by massive-bandwidth connections. Packets sent between cores and processors can include not only data to be processed, but also the OS elements required for processing, and the address for the next processor or core. LambaFabric is low-latency, “super scalable,” and leverages a flow model, rather than traditional “north-south” architectures, to set a new level for computing in data centers and on consumer devices.
As opposed to the huge number of cores connected in parallel in a single-input system like GPUs, KnuPath uses multiple-program multiple-data to enable efficient sparse-matrix calculations, which are more effective than the dense-matrix calculations used by traditional computing systems, according to KnuEdge Chief Marketing Officer Steve Cumings.
KnuEdge believes KnuPath and its LambaFabric will outperform the recently announced Google’s Tensor Processing Unit (TPU), a relatively close competitor. The Google TPU is a chip designed for machine learning, which works with reduced computational precision, to perform operations with fewer transistors.
KnuEdge applied its neural technology approach to voice recognition challenges posed by contacts in the US government and military formed by founder Dan Goldin earlier in his career. Those challenges happen to be very similar to the challenges of providing commercial voice recognition and authentication tools, according to Executive Vice President Kate Dilligan.
“You need to have algorithms that are robust, but that can be implemented in a variety of different ways,” Dilligan told Biometric Update in an interview. “We need to be able to take the crummy audio from a cell phone, from a car’s Bluetooth system, from and iot device, from a contact center . . .you should be able to authenticate.”
Not coincidentally, Internet analyst Mary Meeker’s Internet Trends report for 2016 says “recognition in heavy background noise in far-field & across diverse speaker characteristics” is the “next frontier” in speech recognition.
KnuVerse’ military-grade voice systems brings voice recognition and authentication technology to a level at which it has already begun, through a soft launch, to deliver its security and user experience to companies in finance, health care, and insurance.
One of the first products rolled out by KnuVerse is called Audio PIN.
“It’s prompt obfuscation, so you don’t know what you’re going to be prompted to say,” says Dilligan. “The Audio PIN is providing what you should say in a very specific order, and the algorithm is also very robust. It is able to marry several things together: what you are (using), the PIN, the words, to provide that accuracy of authentication to overcome a lot of the variabilities like a crummy microphone.”
“It works going 70 miles per hour with the top down,” Cumings says, with the enthusiasm of first-hand experience. The “military-grade” solutions come from the challenges posed by the government, so details will be released in pieces. KnuEdge is also planning further product announcements for later this year.
Recently reported announcements of voice authentication adoption include La Banque Postale in March, and UK mobile bank startup Atom in April. Noise is still considered a significant barrier for many applications. If KnuVerse’s noise tolerance is a game changer for voice identification and authentication, the mysteriousness of its stealth period will be addressed with significant attention within and beyond the biometrics industry.