I want to ask, if 30 minutes of data set is a enough to give a decent model? and if I'm making a model of a korean singer can I transcribe the korean lyrics according to their romanization and get good results or would it fail??