생성형 AI 언어모델 프롬프팅 방식에 따른 편곡 가능성 비교 연구

초록

This study analyzes the potential of large language models (LLMs) in musical arrangement. Mozart’s Sonata K.545 was selected as the source material, and two prompting methods—Vanilla prompting and Chain-of-Thought (CoT) prompting—were applied for comparative analysis. The results show that the Vanilla method produced simple chord progressions and lower melodic similarity, whereas the CoT method generated richer harmonies and higher melodic resemblance. These findings demonstrate that prompt design directly affects the creativity and quality of AI-based arrangements. The study suggests that generative AI can be further utilized in music arrangement, game sound design, and interactive content production.

키워드