An adaptive music generation architecture for games based on the deep learning Transformer model

  • Gustavo Amaral PUC-Rio
  • Augusto Baffa PUC-Rio
  • Jean-Pierre Briot LIP6 / CNRS / Sorbonne Université / PUC-Rio
  • Bruno Feijó PUC-Rio
  • Antonio Furtado PUC-Rio

Resumo


This paper presents an architecture for generating music for video games based on the Transformer deep learning model. Our motivation is to be able to customize the generation according to the taste of the player, who can select a corpus of training examples, corresponding to his preferred musical style. The system generates various musical layers, following the standard layering strategy currently used by composers designing video game music. To adapt the music generated to the game play and to the player(s) situation, we are using an arousal-valence model of emotions, in order to control the selection of musical layers. We discuss current limitations and prospects for the future, such as collaborative and interactive control of the musical components.
Palavras-chave: Deep learning, Training, Video games, Emotion recognition, Adaptation models, Music, Games, video game music, adaptive music generation, deep learning, Transformer, layering, emotion model
Publicado
24/10/2022
AMARAL, Gustavo; BAFFA, Augusto; BRIOT, Jean-Pierre; FEIJÓ, Bruno; FURTADO, Antonio. An adaptive music generation architecture for games based on the deep learning Transformer model. In: SIMPÓSIO BRASILEIRO DE JOGOS E ENTRETENIMENTO DIGITAL (SBGAMES), 21. , 2022, Natal/RN. Anais [...]. Porto Alegre: Sociedade Brasileira de Computação, 2022 . p. 13-18.