Don't miss the latest stories
Nvidia Debuts ‘Omniverse Avatars’ To Give AI Assistants Their Corporeal Forms
By Ell Ko, 10 Nov 2021
Subscribe to newsletter
Like us on Facebook
Image via Nvidia
Everyone is racing for their slice of the artificial-intelligence pie, and Nvidia is no exception. On Tuesday, the company revealed the culmination of its efforts in gathering AI technology over the last few months: Omniverse Avatar, a 3D assistant creator.
This takes the likes of Siri and Alexa to the next level. Rather than just being a voice that comes out of a machine from seemingly nowhere in some strange take on a sort of omnipresent modern deity, Nvidia’s new technology will give the voice not only a face, but an entire personhood too, to enhance its presence in our lives.
Avatars created are described as “interactive characters with ray-traced 3D graphics that can see, speak, converse on a wide range of subjects, and understand naturally spoken intent.”
Highly customizable, they will be available 24/7 to help with things like customer service interactions, including making appointments and banking transactions.
“The dawn of intelligent virtual assistants has arrived,” declares Jensen Huang, founder and CEO of Nvidia. “Omniverse Avatar combines NVIDIA’s foundational graphics, simulation and AI technologies to make some of the most complex real-time applications ever created. The use cases of collaborative robots and virtual assistants are incredible and far reaching.”
To demonstrate this new technology, Huang created an avatar in the form of a toy replica of himself. This model then went on to engage various (real) colleagues with conversations on topics like climate science. Its appearance is not reminiscent of a totally realistic person, but it’s not meant to be, either.
Nvidia explains that the avatars’ speech recognition is based on the company’s Riva software, which is a speech recognition tool that spans multiple languages. But its natural language understanding comes from the Megatron 530B large language model, which has a nuanced grasp of human language and can “complete sentences, answer questions of a large domain of subjects, and summarize long, complex stories,” among other things.
Another demo, which also showcases the company’s AI-based toolkit, Project Maxine, shows an avatar being created from a photo of a woman. Riva is used to train a voice for the avatar based on that woman’s own. She can convert text to speech and translate to different languages while sounding largely like a real person.
The avatar can even turn its head while managing to keep natural-looking eye contact, so rest assured it’s not like the doll out of Squid Game with a 360º rotating head and darting eyes.
Omniverse Avatar is part of Nvidia Omniverse, a virtual world simulation and collaboration platform for 3D workflows currently in open beta. For more information, the Nvidia GTC keynote on Omniverse Avatar can be rewatched here, and the GTC event runs through Thursday.
Image via Nvidia
[via Engadget, images via Nvidia]
Receive interesting stories like this one in your inbox
Also check out these recent news