Spectrogram-Based Deep Learning for Flute Audition Assessment and Intelligent Feedback
Manu Agarwal, Ross Greer · 2023
Performers of classical music require a blend of technical precision and artistic expression in their output, and this fusion, often referred to as “musicality,” is considered vital to performance quality. This paper introduces an innovative approach that leverages deep learning and LLMs to simultaneously evaluate and coach musicians using recorded performances on a variety of performance metrics at varying levels of subjectivity. A case study, centered around flute players performing a challenging excerpt from Ravel’s “Daphnis et Chloé,” demonstrates the proposed model’s capabilities. Feedback is generated by a large-language model based on machine-assessed quality, learned from human judgments. The model showcases promise in bridging the gap between technical precision and human expression in classical music performance assessment and provides a foundation for expanding the repertoire of assessed pieces and advancing the integration of AI in classical music education.