Any documentations on training from scratch using custom data in other languages ?

I have a speaker diarization dataset in Vietnamese, where, in every audio file, segments of speakers are already annotated. How should I prepare and process data to be able to train UIS-RNN on my custom data ?