In most github repositories for machine learning based text to speech, the LJ-Speech dataset is being used and optimized for.
Having unsucessfully tried to use my own wave files for it, I am interested in the right approach to prepare your dataset for an optimized framework to likely convert.