KnuEdge, a company specializing in cloud-based voice recognition technology announced its KnuVerse voice authentication solution that can work in any type of noisy environment, as well as its neural KnuPath chip architecture for voice recognition.
ASIC is an implementation method - not an architecture. The same design could be implemented on a FPGA, if it will fit.
256 GFLOPS in 34 W doesn't sound particularly impressive. Maybe they had trouble finding mil-spec parts that were fast & efficient enough, but it seems like both Movidius and Tegra TX-1 are more efficient.
Neither has 40 GB/sec of PCIe, though. It's kinda hard to see why they'd need so much I/O bandwidth, unless you need to interconnect a bunch of these to do anything useful. In that case, I'd have to ask: why not just use a bigger GPU?
They seem to be misinformed. Whether voice or fingerprints, biometrics should not be activated if you are security-conscious.
It is now getting known that the authentication by biometrics usually comes with poorer security than PIN/password-only authentication. The following video explains how biomerics makes a backdoor to password-protected information. https://youtu.be/5e2oHZccMe4