It’s not a generative model running on your graphics card coming up with novel speech to match the gamestate; it would never be fast enough (and you’re also using that card to, you know, run the graphics). They’re using machine learning to generate speech for their pre-written lines so that they can avoid hiring voice actors. I guess they didn’t want to try the old way of having whoever isn’t too busy in your office record the lines.
It’s not a generative model running on your graphics card coming up with novel speech to match the gamestate; it would never be fast enough (and you’re also using that card to, you know, run the graphics). They’re using machine learning to generate speech for their pre-written lines so that they can avoid hiring voice actors. I guess they didn’t want to try the old way of having whoever isn’t too busy in your office record the lines.
Depending on how many variables/contexts there are in the lines, that could still be a combinatorial nightmare to record.