3 ms·

xtts2 is great, but it looks like this model is probably more consistent with its output and has a better grasp of meaning in long texts.