Roar is a transformer-based text-to-audio model created by Alola. Currently, two checkpoints are supported: a small and a large version.