[Feature Request] LLM-Based Text Cleanup

[realThirdpartycookie]

It would be cool if text would be sanitized chunkwise before being fed into the TTS. Like with GPT-4o or Ollama.

[ROBERT-MCDOWELL]

it's widely sanitized already by chapter or "block" of text. we cannot add more and more LLM or A.I. on TTS A.I. which sometimes needs more than 16GB or RAM with 3 active models. The installation is already huge, so integrate more LLM makes the maintenance (packages version etc..) almost impossible.

[DrewThomasson]

Thanks for the suggestion,

BUT, I have already experimented with this idea in a diff repo, and the results were not to my liking.

https://github.com/DrewThomasson/ext-Reformatter-LLM

As magic as they seem, they give inconsistent results, are EXTREMELY resource intensive, will cause a bunch of bugs etc...

[realThirdpartycookie]

That is unfortunate. I mostly use it for scientific literature and it's already difficult to read for a human.
I wish to remove inline citations, appendices, artifacts from tables and so on.

[ROBERT-MCDOWELL]

we have planned to make a pre validation of the conversion, showing a modal where you ll be free to remove/select the detect text block and modify its content before the conversion.