Algorithmic interactive music generation in videogames

A modular design for adaptive automatic music scoring

Abstract

In this article, I review the concept of algorithmic generative and interactive music and discuss the advantages and challenges of its implementation in videogames. Excessive repetition caused by low interactivity in music sequences through gameplay has been tackled primarily by using random or sequential containers, coupled with overlapping rules and adaptive mix parameters, as demonstrated in the Dynamic Music Units in Audiokinetic’s Wwise middleware. This approach provides a higher variety through re-combinatorial properties of music tracks and also a responsive and interactive music stream. However, it mainly uses prerecorded music sequences that reappear and are easy to recognize throughout gameplay. Generative principles such as single-seed design have been occasionally applied in game music scoring to generate material. Some of them are complemented with rules and are assigned to sections with low emotional requirements, but support for real-time interaction in gameplay situations, although desirable, is rarely found.
While algorithmic note-by-note generation can offer interactive flexibility and infinite diversity, it poses significant challenges such as achieving human-like performativity and producing a distinctive narrative style through measurable parameters or program arguments. Starting with music generation, I examine conceptual implementations and technical challenges of algorithmic composition studies that use Markov models, a-life/evolutionary music, generative grammars, agents, and artificial neural networks/deep learning. For each model, I evaluate rule-based strategies for interactive music transformation using parameters provided by contextual gameplay situations. Finally, I propose a compositional tool design based in modular instances of algorithmic music generation, featuring stylistic interactive control in connection with an audio engine rendering system.