If the celebs are just doing pre-scripted videos with fill in the blanks, how hard would it be to alter the audio and fill in the blanks yourself using a machine?

IIRC there was a Google deep AI project or something similar which was able to "match" a human voice.