Hacker News Reader

Not convinced. The second reference has a slight Indian accent that isn’t carried over into the generated samples.

Training data bias?

What are you talking about? I am not noticing it at all.