“isolate little snippets of sound called phonemes, and they associate multiple keywords with each one. You end up with a database of little digital fragments of sound, each one distinctive. When you want to make the voice speak, you feed another script into the cloning system, and the algorithms do a keyword lookup and”