Consider the following VoCo project from Adobe.
In it they show how they are able to take 20 mins of speech and create additional narration in the same voice very accurately.
Is there a neural net or something that is available in the language that can produce similar results.