Just imagine private model (own, little spech-to-text model) listening to the people playing DnD feeding to the bigger brother that in turn controls the music generation and feeds it back to the speakers as the story progresses.

:)