NVIDIA’s ‘Audio2Face’ AI Creates A Face Model To Speak The Words You’re Saying

Video screenshot via NVIDIA

Are you camera shy? Is public speaking not your forte? One day, you might be able to get a 3D avatar to speak in your stead.

This week, NVIDIA introduced the beta version of its cloud platform for 3D design called ‘Omniverse’ and with it came ‘Audio2Face’, a software that generates a photorealistic 3D face to realistically convey your speech in real time.

The avatar’s appearance is created by the user, and it will then move its “muscles” as if the words had come out of its own mouth.

This tool will be a boon for animators, who will no longer have to manually contort the mouth area based on what a character is meant to say.



For the general public, the 3D model can even serve as a stand-in for when the internet connection in a video call is stalled and your video isn’t moving, according to Gizmodo Australia.

‘Audio2Face’ is still in a preliminary stage, but NVIDIA expects that the full version will be ready in early 2020.





[via Gizmodo Australia, cover image via NVIDIA]