Source: Nvidia/YouTube.
Dialogue choices are a very important narrative tool for many video games, but the detail is that the options offered and the tone implied when selected often don’t match exactly. Generative AI is all the rage in tech right now, so the natural question is, what would happen if we added that technology to our games? As part of his Nvidia showcase at COMPUTEX 2023, the company showcased what it called his ACE (Avatar Cloud Engine) For Games.
The company describes Nvidia ACE For Games itself as a middleware suite that includes the generative AI toolkit NeMo, an automatic speech recognition (ASR) and text-to-speech (TTS) combo known as Riva. Best of all is the Omniverse Audio2Face technology that matches NPC facial expressions to spoken dialogue. The company specifically mentions Unreal Engine 5 compatibility with his creator of MetaHuman characters.
There’s even a demo video showing Nvidia ACE For Games in action, but to be fair, the scripted showcase is not a show off of the potential of generative AI. In fact, aside from the apparent voice input of the player’s character dialogue, the dialogue is not much different from dialogue choices in modern games.
But it’s the imperfections that this film shows, especially for those who are picky about details. According to the company, the technology enables “accurate facial animations and expressions, all in your native language.” In the case of this showcase video, the name Kumon Aoki is often overpronounced, as the mother tongue is clearly American English.
According to The Verge, Nvidia’s vice president of GeForce platforms, Jason Paul, said ACE For Games could expand to multiple characters at once, allowing NPCs to talk to each other. In theory anyway, since Paul says he’s never actually seen it tested either. This is great for immersing yourself in an RPG with a huge world, but you can already imagine it triggering an uncanny valley feeling.
(Source: Nvidia, The Verge)
