I was researching for really good open source text to speech programs yesterday and I ran into this project by Mozilla. By the way here is the best voice output I’ve heard from it [1].
If you want to run the prebuilt models you can use the colab Jupiter notebooks that user tugstugi put up on this page [1]. The voice from the comment above comes from WaveRNN and Tacotron2.
[1]: https://m.soundcloud.com/user-565970875/ljspeech-logistic-wa...