Wave Function Collapse for Videogame Music Generation

This thesis project falls into a topic that gained increasing popularity in these years, which is computer creativity, regarding the use of AI techniques to generate multimedia art forms. This topic in particular is receiving particular attention from the video games industry, as oftentimes AI is used for generating assets inside the game engine itself, such as graphics and textures. Several useful tools are used to efficiently perform this task, such as Procedural Content Generation (PCG). However, Procedural Content Generation has not really been used for sound multimedia, such as audio, sound, or music generation. This thesis wants to explore an alternative method, by adapting an already existing tool (Wave Function Collapse), used for PCG, to automatically generate a soundtrack for a game. In the first section of the thesis, a brief State of the Art paragraph will describe the current state of the research on this topic and the reasons as per why this method could be useful in the future. Then the thesis will continue by describing the prototype that was written in Python language, in order to have a simple benchmark to test out the overall feasibility of the idea. The prototype allows a human to write simple original musical pieces in MIDI notation, and then to ‘feed’ them to the WFC algorithm. The output will be a song that tries to be musically cohesive to the original input musical file. The thesis concludes by reporting some observations on the results obtained and potential future developments. In conclusion, one of the aims of the project of this thesis is to adapt an already existing PCG technique used for game textures into a tool for generating music.