Yeah, and that's my understanding. Nothing goes video -> text, or audio -> text, or even text -> text without first going through state space. That's where the core of the transformer architecture is.