Copy paste into an LLM, piece by piece; ask it to keep the text the same but strip out the crap you don’t need. Give it specific examples of what things you want it to strip out. Then feed that output to the text-to-speech.
My recommendations in order of which to try first would be:
ChatGPT free tier
Claude.ai $20/mo tier
ChatGPT $20/mo tier
Re-give the instructions with every page; it won’t remember what to do. Honestly you probably want to start a new chat for each page / 2 pages / 5 pages or whatever. With simple instructions like this it can handle quite a lot of text at once so I wouldn’t be shy about how much you paste unless you start to hit limits or something.