For TTS I wonder how this compares to nanotts[1] with the en-GB voice, which is sort of unreasonably good.

[1] https://github.com/gmn/nanotts