Indie game developers typically handle every aspect of their game themselves, from coding to visual assets. However, music and sound effects are usually outsourced to third-party producers or sourced from online resources (paid or open). While outsourcing audio production might be a viable option for established indie developers, it’s less accessible for those with hard budget constraints, particularly in developing countries. To make music production for games more democratic, we are investigating neural models for learning to generate soundtracks for video games directly from gameplay videos.

Deep Learning models have shown very promising results in automatically composing polyphonic music pieces. However, it is very hard to control such models in order to guide the compositions towards a desired goal. In this project, I am interested in controlling deep learning model to automatically generate music with a given sentiment or emotion.

Physics-based puzzle games add an extra layer of difficulty to procedural content generation (PCG) because their mechanics are based on 'realistic' physics. Thus, evaluating feasibility and quality is harder and typically requires simulations. This project consists of designing PCG methods capable of generating feasible and interesting levels for physics-based puzzle games, such as Angry Birds.