Once someone gets an implementation of https://sample-efficient-adaptive-tts.github.io/demo/ [0] working that should be good enough (quality-wise).

[0] https://arxiv.org/abs/1809.10460