Computer-generated music can provide interesting insights into the structure of music and serve as inspiration for novice and professional composers alike. I investigate the use of the Transformer-XL neural network architecture for interactive co-creation of symbolic music in the style of pop piano. I present a modular system consisting of two software components: backend (music generation engine) and frontend (user interaction). I evaluate the neural network architecture and discuss the overall system with regard to higher-level issues in the field of computational creativity. Based on musical prompts, the system can be used to iteratively generate musical pieces of several bars length. However, it does not generalise well to new data, hindering interaction with complex user prompts.
Music generated from scratch:
Music generated from prompts: