Scholarlink

<p>Piano instruction has traditionally depended on one-on-one human interaction, limiting its scalability and adaptability for diverse learners. The integration of artificial intelligence into music education introduces opportunities for more effective, data-driven teaching strategies. Deep learning methods are capable of modeling musical features, including accuracy, timing, and expressive dynamics. This study proposes a Deep Learning-based big data model to Enhance Piano Teaching (DLEPT), a system designed to improve piano instruction through real-time, multimodal assessment and intelligent feedback generation. DLEPT uniquely combines MIDI and audio inputs to provide expressive, context-aware performance analysis and generates pedagogically-informed corrective feedback. DLEPT is developed using the MAESTRO dataset, which contains over 200&#xa0;h of professionally performed and aligned MIDI and audio recordings. A Transformer-based architecture is employed to compare student performances against expert references, detecting deviations in rhythm, dynamics, and phrasing. The model is optimized using the AdamW optimizer, combined with learning rate warm-up and cosine decay scheduling, with additional regularization such as dropout and label smoothing, to ensure stable convergence and generalization during training. Corrective feedback is then generated to guide learners in improving technical accuracy and musical expression. The trained system achieved an accuracy of over 92% in detecting common performance issues and produced feedback validated against expert-level evaluations. The combined use of MIDI and audio inputs enabled expressive and context-aware performance analysis. In conclusion, DLEPT provides a scalable and intelligent solution for enhancing piano instruction through modern deep learning and optimization techniques, representing a novel multimodal approach in intelligent music pedagogy.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Strategies for improving piano teaching effectiveness using deep learning with big data models

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年7月21日

查看全文
  • Huaman Gao

摘要

Piano instruction has traditionally depended on one-on-one human interaction, limiting its scalability and adaptability for diverse learners. The integration of artificial intelligence into music education introduces opportunities for more effective, data-driven teaching strategies. Deep learning methods are capable of modeling musical features, including accuracy, timing, and expressive dynamics. This study proposes a Deep Learning-based big data model to Enhance Piano Teaching (DLEPT), a system designed to improve piano instruction through real-time, multimodal assessment and intelligent feedback generation. DLEPT uniquely combines MIDI and audio inputs to provide expressive, context-aware performance analysis and generates pedagogically-informed corrective feedback. DLEPT is developed using the MAESTRO dataset, which contains over 200 h of professionally performed and aligned MIDI and audio recordings. A Transformer-based architecture is employed to compare student performances against expert references, detecting deviations in rhythm, dynamics, and phrasing. The model is optimized using the AdamW optimizer, combined with learning rate warm-up and cosine decay scheduling, with additional regularization such as dropout and label smoothing, to ensure stable convergence and generalization during training. Corrective feedback is then generated to guide learners in improving technical accuracy and musical expression. The trained system achieved an accuracy of over 92% in detecting common performance issues and produced feedback validated against expert-level evaluations. The combined use of MIDI and audio inputs enabled expressive and context-aware performance analysis. In conclusion, DLEPT provides a scalable and intelligent solution for enhancing piano instruction through modern deep learning and optimization techniques, representing a novel multimodal approach in intelligent music pedagogy.