Former NASA Administrator Unveils Voice Recognition, Chip Designby
Dan Goldin’s KnuEdge aims to build neural computing networks
Company is backed by $100 million in anonymous angel money
Dan Goldin is no stranger to launches. He oversaw 62 Space Shuttle liftoffs as the longest-serving administrator of the National Aeronautics and Space Administration. On Monday he presided over an entirely different sort of launch: two new products -- a voice authentication technology and a neural computing chip architecture -- following a decade of secretive research and development by his San Diego-based company, KnuEdge.
Backed by $100 million from angel investors, KnuEdge is working on neural computing solutions, including hardware and software for machine learning. Voice recognition is the first problem the company has tried to address.
KnuEdge’s voice authentication, which is being tested by several businesses including XL Catlin, the insurance and reinsurance company that’s part of Dublin-based XL Group, can recognize an individual’s voice even in extremely noisy environments that would stump most existing software, Goldin said.
"Siri, Cortana, Alexa -- these are all great products but you get a little noise in the background and they don’t work," he said, referring to the popular digital assistants, which are powered by natural language processing and voice recognition, built by Apple Inc., Microsoft Corp. and Amazon.com Inc., respectively.
Goldin said KnuEdge’s voice authentication, called KnuVerse, has been successfully tested in situations where there are other voices talking and in cases where the extraneous noise was as much as eight times more powerful than the speaker’s voice. KnuEdge also hired four of the world’s top voice impersonators to try to fool the system. “They couldn’t do it,” Goldin said. “It wasn’t even close.”
While Goldin said this voice recognition technology will have applications in all kinds of human-computer interaction, including digital assistants and chatbots, KnuEdge decided to start with a security application -- voice authentication -- because it’s an important issue across industries.
Founded in 2005, KnuEdge’s other product is a computer chip with a 256-core unit called KnuPath. A new neural computing architecture unveiled Monday, called KnuPath LambdaFabric, will allow as many as 512,000 of these chips to be connected, providing super fast computing speeds with far greater energy efficiency than is currently possible, according to the company. Goldin said that this type of architecture -- which is based on insights from studying the way the human brain works -- are ideally suited to machine-learning.
Existing chips are usually based on an architecture pioneered by computer scientist John von Neumann in which data and programs are held in memory and the processor is a separate area of the chip that moves data from memory over a electronic pathway called a bus. The bus’s capacity limits processing speed and KnuEdge’s architecture is designed to break the bottleneck. IBM and other companies are also working on technology to solve this problem.