GPT based DF speech generator project is ON

[Maverick_Y]

The self-made map is highly welcome in DF communication. However, the authors of most maps are only able to use the official voices present in the game, which greatly limits the use of voices. Very often we have to use wrong unit names （especially for BHD Blackhawk） and wrong background descriptions to partially express the intention of the mapper.

With the development of AI and Large Language Modelling (LLM), I see the possibility of using LLM to generate DF-like speech. The clean, noise-free sound source files provided in the game, as well as accurate subtitle labelling, can serve as very good training material. I would like to train a language model that makes it very simple for every map author to output a DF-style speech file after typing in the text.

Once it has been down, I will update all my maps with completely new audio.

[Oscarmike247]

If it sounds right, this would be very useful!