I just tried out the new Supertonic TTS for 64-bit NVDA, and I was more impressed with it than I thought. It is a local neural TTS, with as of right now, 10 voices that are English only, , (five female, five male, four American and one British per gender), and my favorite is F3. What I find impressive is that when the speech quality control value is at its lowest, it still sounds decent, though due to the nature of neural TTS engines, just like Sonata, the longer the paragraph, the longer the delay.