Comment by cellis

Could this be used to train a text -> audio model? I'm thinking of an architecture that uses RVQ. Would RVQ still be necessary?