This page accompanies the report Génération automatique de jingles audio. Please refer to this document for context.
Warning: some sounds can be loud.
Our best model was trained for 2 M steps (231.6 M learnable parameters). We partially set the control to -1 (unconditional).