Training dataset

[hahust191806]

Hi guys!
I'm doing a music inpainting problem, my input is a vocal track, my task is to generate a music segment corresponding to that vocal, so my dataset includes 2 parts, 1 is vocal, 2 is instrument. So what will be the dataset for the Semantic, Coarse, Fine phase in the two vocal and instrumental episodes?

[LWprogramming]

i didnt try a ton of things or get great results with the approach i tried when experimenting with this but maybe it'd work to glue the vocal and instrumental parts separated by special tokens (in my case i just inserted a totally silent segment lol) and hope the transformers can learn to predict the accompaniment from the melody