A Study of Artificial Intelligence-Based Emotional Music Generation Systems

Artificial intelligence-based affective music generation (AI-AMG) systems generate music automatically with the purpose of expressing or evoking particular feelings. This review examines studies published between 1990 and 2022 and organizes existing systems according to their main components and generation methods. A typical AI-AMG system includes target emotion identification, emotional music generation, and emotional evaluation. Rule-based, data-driven, optimization-based, and hybrid are the four categories of generation methodologies. Rule-based systems include connections between musical features and emotions, whereas data-driven systems identify structural patterns using Markov models or neural networks. Optimization methods use techniques such as genetic algorithms and graph or tree traversal, and hybrid systems combine multiple approaches. Evaluation is commonly performed through computational measures or human listener studies that assess perceived or induced emotion. Current systems can produce music associated with target emotional states, but important limitations remain, including dependence on labeled datasets, inconsistent evaluation procedures, limited measurement of induced emotion, and lower creativity than human composition. The review highlights the need for more reliable evaluation, improved emotional control, and systems that generate natural, diverse, and context-sensitive music.