selfcontrol7/Korean_Voice_Phishing_Detection ? reverse-engineered prompt

Build me a Korean voice phishing detection project that can run the main experiments from this repo and make it easy to compare the different approaches in one place.

I want the code to cover text only models, hybrid attention based models, KoBERT based models, federated learning experiments, and the newer multimodal setup that combines audio and text. Please make it easy to preprocess the call data, train the models, evaluate them, and show the results clearly, including call level predictions when that matters.

If the data or paper specific files need a bit of cleanup to run smoothly, please wire that up too and make the notebooks or scripts practical to use from start to finish. I’d also like a simple way to reproduce the plots or summary tables from the experiments.

Use the repo structure that is already there, and look up current docs online if you need to.