AI-driven emotional intelligence in piano education: Deep learning models for expressive performance coaching

With the increasing use of artificial intelligence (AI) in higher music education, there is a growing need to evaluate whether intelligent feedback systems can meaningfully enhance expressive piano performance. This study investigates the research question: Does the use of an AI-based analytical performance assessment system lead to greater improvements in piano expressivity than traditional instruction? To address this question, we developed an intelligent performance-assessment system that applies modern BiLSTM attention technologies to analyze expressive qualities of piano playing, including dynamics, phrasing, and emotional conveyance. A total of 290 Chinese undergraduate piano students (aged 18–22) were randomly assigned to an AI-supported group ( n = 145) or a traditional-instruction control group ( n = 145) for a three-month training period. Expressivity was measured using a hybrid 100-point scoring system that integrated objective performance analytics with expert double-blind evaluations. Statistical analysis included paired and independent t -tests and effect-size estimation. The intervention group demonstrated a substantial improvement in expressive performance (Δ M = 12.97; t (144) = −21.64; p < .001; d = 1.80), indicating a large effect. By contrast, changes in the control group were minimal and statistically nonsignificant ( p = .092; d = 0.14). These findings provide strong empirical evidence that AI-based feedback can significantly enhance students' emotional expressivity in piano performance. • A deep learning model was built to process MIDI and audio files with modern methods • The intervention group showed a significant increase in a mean expressivity score • DLM-based method is effective in enhancing the emotional expression of piano playing • AI can effectively support personalized training and enhance music pedagogy