[Feature Request] Handling contractions, (don't, designer's) Handling hyphens (zero-shot, end-game), unusual names or word pronunciations

First, this is an absolutely amazing repo. Thank you for making it available.

I've been preprocessing text by removing contractions (don't → dont) and hyphens (zero-shot → zero shot) to improve generation quality, but some issues still slip through. I'm curious if there are documented best practices for text preparation. I haven't found anything in the wiki yet. Feature requests:

1. Phonetic override system: A way to define pronunciation for proper names and unusual words on a per-project basis would be incredibly valuable.
2. Improved contraction handling: Some contractions work well (isn't, aren't), while others are inconsistent. For example "don't" is inconsistent sometimes producing "dohn tee" and sometimes producing "don't". However "designer's" always becomes "designer ess". So I write it designers to fool it into saying it correctly.
3. Structural awareness: The system doesn't recognize paragraph breaks or heading context. Chapter/section titles without terminal punctuation run directly into the following text:

   1.1 The Designers Dilemma
   In landscape architecture, every creation carries an unspoken cost.

Renders as: one point one the designer ess dilemma in landscape architecture, every creation carries an unspoken cost. rather than one point one. the designers dilemma. in landscape architecture, every creation carries an unspoken cost.

As previously stated, I'm currently working around these by editing a preprocessed version of my text, which works fine. Just wondering if handling paragraph structure and heading context is on the roadmap. A method for handling phonetics for unusual names and unusual terms (like medical or legal terms) would also be useful.