With the rapid development of artificial intelligence, multiple specialised machine learning tools have emerged, capable of generating photos, 3D models and music. These models can accurately generate new pieces of media but fall short when being used to help fuel the creative process. From the perspective of an artist, an AI-powered tool can take away from their own creative expression, especially within the realm of music, where small flairs in notes, timing and dynamics can separate individual approaches. There currently is a gap where an artist may want to generate a small section of music based on a previous sample of their own music, whether it be to experiment or to overcome writer's block.

Out of this forms a problem: creating a tool for artists to experiment with generating music based on a sample of their own music. I will specifically focus on a single track sample from a single song, which, whilst being a reduction in scope, should still result in a useful creative tool.

I intend to approach this problem algorithmically by utilising forms of computational optimisation. By analysing a section of MIDI music and parsing it into pitch classes and time-point intervals alongside other structural elements before applying a genetic algorithm to generate a new section of music based on the input. The program will also use the same parsing to analyse and compare the generated section to the inputs.

I consider this project to be sufficiently challenging, as designing a genetic algorithm that can generate a distinct yet coherent piece of music would provide a significant hurdle to overcome, especially if I were to design adjustment sliders for genres or moods.