There are primitive ways to alter mouth movements but the quality is subpar for me to consider using it at the moment.
You could film yourself acting the role, rotoscope yourself onto the original actor, then apply a deep fake on top of all that.
I’m assuming that deepfakes use some kind of automated facial tracking to keep the deepfake in the correct place, angle, and lighting - if that’s right, would that same technology be able to be used in the rotoscoping stage, to seamlessly line up a new facial performance?
My reason for being so curious about this is pretty obvious for fanediting - imagine the scope that would open up if a faneditor could completely replace an entire character’s face and dialogue with their own new actor.