GenerationMania: Learning to Semantically Choreograph

Abstract

Beatmania is a rhythm action game where players must reproduce some of the sounds of a song by pressing specific controller buttons at the correct time. In this paper we investigate the use of deep neural networks to automatically create game stages—called charts—for arbitrary pieces of music. Our technique uses a multi-layer feed-forward network trained on sound sequence summary statistics to predict which sounds in the music are to be played by the player and which will play automatically. We use another neural network along with rules to determine which controls should be mapped to which sounds. We evaluated our system on the ability to reconstruct charts in a held-out test set, achieving an F1-score that significantly beats LSTM baselines.

Automatic generation of stage data for music games with sparse target density

“…Lin et al [10] employed the concept called ''beat phase'' which is a similar concept to our method in that it uses musicology knowledge. Beat phase denotes the position of the target in relation to the temporal position between two beats.…”

Maimai Chart Generation Model Based on Machine Learning Algorithms

“…For instance, DDC utilized two separate machine learning models to solve the tasks of placing notes and note selection [2]. The chart generation model proposed for Maimai, however, resembles those utilized in GenerationMania and Beatmania IIDX [10]. In a more traditional rhythm game, the key players' input will directly influence the song.…”

A Demonstration of KiaiTime: A Mixed-Initiative PCGML Rhythm Game Editor

“…There have been a number of PCGML systems created for the task of chart generation for various rhythm games aside from our chosen approach, TaikoNation (Halina and Guzdial 2021). The majority of these prior works focus on subproblems in chart generation which are irrelevant to our editor, like sample classification (Lin, Xiao, and Riedl 2019) and improved onset detection (Liang, Li, and Ikeda 2019). The most general and applicable other choice of chart generation system utilizing PCGML is Donahue's Dance Dance Convolution (Donahue, Lipton, and McAuley 2017).…”

“…There has been prior work on automated chart creation (Donahue, Lipton, and McAuley 2017;Lin, Xiao, and Riedl 2019;Liang, Li, and Ikeda 2019), the most recent of which utilize procedural content generation via machine learning approaches (PCGML) (Summerville et al 2018). However, none of these systems include mixed-initiative co-creative functions beyond the selection of difficulty of a generated chart.…”