He built a hit podcast about the Epstein files. It's AI-generated
fastcompany.com
fastcompany.com
In the past, when I used the local KokoroTTS model, I did something a bit more primitive for longer texts where it constantly mispronounced a word: a pre-TTS pass with regex to replace those words with a phonetic-sounding equivalent that worked more consistently (e.g. replacing “Danish” with “Day-nish.”)
I have a similar setup - I let it spin in the background on some various RSS news feeds that I can listen to while I go walking.
If you don't need voice cloning support it's one of the best locally hostable ones out there IMHO.