ABSTRACT

This study examines the characteristics of human evaluation in secondary school music performance assessments and explores how these school-based tendencies can inform the development of AI algorithmic models. Although prior research has noted the influence of intuitive and holistic impressions in music performance evaluation, such insights have rarely been connected to the realities of assessing non-major middle school students in authentic school settings. In this study, music-major evaluators assessed actual middle-school vocal and instrumental performance recordings collected during the development of an AI assessment platform, and subsequently participated in in-depth interviews with AI developers. The analysis showed that evaluators placed particular importance on students’ sincerity, effort, and engagement, and that expressive elements and pitch accuracy required flexible, context-dependent interpretation rather than strict precision. These tendencies highlighted the difficulty of converting school-based evaluative practices into forms that AI can learn reliably. By situating human evaluation patterns within the contextual and pedagogical aims of school environments, this study offers insight into how AI systems can be designed to align with the performance characteristics of non-major learners and emphasizes the need for human-informed, context-sensitive AI models in music education.

Keyword :