Modified interactive genetic algorithm system for music gene-ration

• Date: 2026
• Main Authors: Федорченко, Є. М., Олійник, А. О., Федорончак, Т. В., Степаненко, О. О., Чорнобук, М. О., Зайко, Т. А., Федорченко, Ю. В.
• Format: Article
• Language: Ukrainian
• Published: Інститут проблем реєстрації інформації НАН України 2026
• Subjects: MIDI; genetic algorithm; music generation; machine learning
• Online Access: https://drsp.ipri.kiev.ua/article/view/363158
• Journal Title: Data Recording, Storage & Processing

Dynamic music generation is a mature field of digital audio processing and is commonly applied in interactive multimedia systems such as games, multimedia applications and adaptive audio systems. These systems produce musical accompaniment that adapts to changes in the environment, interaction context or user attributes, which boosts audio diversity and enhances user experience. Artificial intelligence (AI)-driven music generation has seen rapid advancement in recent years. Current systems can generate instrumental and vocal music from textual descriptions or style information and can be perceptually high quality. Unfortunately, the systems are often not suitable for use due to high computational effort, resource demands and low compatibility for real-time deployment on user devices. This motivates a need for alternative approaches that can generate music locally or on the fly with reduced resource demands. Such an approach is interactive genetic algorithms, in which the fitness function is determined by the user. Rather than employing other optimization methods with well-defined objective functions, these systems use the user’s preference to drive evolution of musical content. They are particularly well suited for applications in music generation, where the quality of the output is subjective. Current evolutionary music systems show the relevance of such techniques. At the same time, there are still many systems suffer from user fatigue, lack of convergence and poor adaptability of evolutionary parameters. In this work, we present an adaptive interactive genetic algorithm for generating MIDI music. The approach is still computationally lightweight, and it records user feedback over the generations and estimates user engagement based on statistics calculated from the rating stream. Using this prediction, the system adaptively controls the probabilities of mutation and injection to alleviate user fatigue, hasten convergence and keep the user engaged during the generation process. The proposed method is lightweight, adaptive, and is designed to generate music in resource-limited environments in real time. Tabl.: 2. Fig.:4. Refs: 12 titles.