Show HN: I built a smart audiobook converter that sounds human (not KokoroTTS)

Hi again!

Existing text to speech or ebook to audiobook apps usually rely on KokoroTTS and don't process the pdf in a reliable way, so it often ends up reading all the front matter/opening section OCR artifacts and other crap in a machine-sounding voice. My app first fixes the content structure using an LLM and then uses heavier models to get voices with natural intonation across more than 40 languages.

About a year ago when I posted this it was still just an idea, now its becoming a real thing. Any feedback appreciated.