How expensive would it be to build a prototype for something like this? http://www.dailymail.co.uk/sciencetech/article-5630775/Viral-deepfakes-video-Barack-Oba

How expensive would it be to build a prototype for something like this? http://www.dailymail.co.uk/sciencetech/article-5630775/Viral-deepfakes-video-Barack-Obama-shows-calling-President-Donald-Trump-total-dips.html http://www.dailymail.co.uk/sciencetech/article-5630775/Viral... but for audio? Assume I have the legal rights from the actor secured. How much video data would I need? Could I use something like Google's youtube dataset as a starting point? Could you build some kind of eigenfaces only for voices? To generate some kind of generic voice and then feed it into some kind of model of an actor to get their voice?