Yes but this stuff is taking 100 gigabytes per game

Once we get flexgen running Llama (or some other combination of optimizations) there will be no audio files and these LLM’s can be run client side on consumer hardware

probably as a shared resource at the OS level

Pre-programmed common stuff sure which many games do as they can tell where you are, vehicle in and any weapons used etc, steal a (police car|military jet|ambulance) etc and they can craft scenarios using audio for those but for more natural random somewhat unpredictable stuff you would have endless combinations that need to be accounted for. It would have to act in response to a bunch of actions so not feasible.

A good example is Mortal Kombat where each character has several lines for every OTHER character (https://www.youtube.com/watch?v=L85QApISlvA) they are about to fight, that's a LOT of voice work as opposed to using the character lore and their history and or relationship with the opponent to come up with something fresh that's (witty|snarky|sad) to say etc.