Co-creative Orchestration of Angeles with Layer Scores and Orchestration Plans

Orchestration is the process of creating music for a group of instruments, combining, blending, and contrasting their sounds to produce a unique orchestral texture. In this research/creation project, our team was commissioned to create an AI/human orchestration of two movements of Angeles, a piano composition by Gissel Velarde. The project turned out as a perfect case study for computational creativity in music and orchestration, where the role of the model is between AI as a colleague and AI as a tool. Our main contribution is a preliminary framework for computer assisted orchestration. By modeling a layer score and an orchestration plan in the orchestration process, we implement a simple Markov model that selects possible instrumentations for each score segment. Personalization of the AI and AI/human interaction occur through human segmentation of the score at two stages of the process (layer score, orchestral segments with loudness profile), through instrumentation presets, and finally through selection of the final orchestral plan and through the actual orchestration. We detail the research aspects of this co-creative project and analyze the roles of the actors involved in the creation of the final piece: the Music Information Retrieval (MIR) researchers, the orchestrators, and the algorithms.