The abstract from the paper is the following:
We introduce Jukebox, a model that generates music with singing in the raw audio domain.