A Systematic Overview on AI Music Generation, Arrangement, and Orchestration

Abstract

The rapid growth of AI-based music generation, arrangement, and orchestration has resulted in a large and fragmented body of literature, making it increasingly difficult for researchers and practitioners to navigate recent developments and identify the most promising approaches. Existing surveys often focus on specific subfields or do not fully reflect recent advances such as diffusion models, transformer-based architectures, and large-scale commercial systems. This thesis addresses these gaps through a systematic literature review reported with reference to the PRISMA 2020 guidelines, investigating AI methods for music generation, arrangement, and orchestration published between 2020 and 2025. From 4,521 records identified across six academic sources, 1,083 could be exported to Zotero for record management and screening; after deduplication and eligibility assessment, 107 primary studies were included in the review. For each included study, information regarding methodology, datasets, musical representations, evaluation procedures, performance metrics, and code availability was extracted and organised in a comprehensive comparison table. The review extends the taxonomy proposed by Zhu et al. (2023), synthesises eight major research trends, and identifies and justifies 27 systems for in-depth analysis, whose available demonstrations and outputs are examined through a pilot exploratory listening analysis, and whose author-reported evaluations are synthesised separately. The contributions include an updated taxonomy of AI music systems, a structured comparison of 107 studies, a trend analysis spanning 2020–2025, and practical recommendations for future research in AI-assisted music generation, arrangement, and orchestration.