Papercup uses machine learning to train its text-to-speech platform to isolate the voices in a video, translate them into the target language, and create a synthetic voiceover that sounds like a human speaking in the target native language.