When you work with models, you have to train them using real-world data. Artificially-generated data isn't of some help.

The difference between a TTS speaker and a human speaker is relevant and can be discerned easily (even by human ears).