Abstract
At the heart of many rhythm-based games are “charts”, which align music to actions sought from the player. Charts are often time-consuming to make, as they require a deep understanding of game mechanics and the music. We describe GORST, a system that generates initial suggestions for a chart given a piece of music. GORST chops music into beats, which are encoded and fed into transformer models four beats at a time to output chart tokens”. The output tokens are then used as prompts in the next round of generation. Our evaluation and inspection shows that GORST outputs very plausible charts.