Automated Pop-to-A Cappella Score Generation

A significant number of research has explored the co-creation and automated generation of symbolic music across various genres. In this work, we investigate the potential for a comprehensive, automated system that transforms polyphonic pop music into six-part a cappella sheet music, featuring a lead vocal line, SATB harmonies, and vocal percussion. Leveraging advanced AI techniques—including pre-trained models and custom algorithms—the system is designed to tackle challenges in music transcription, feature extraction, and SATB harmonization. Our approach centers on curating paired datasets that align solo piano MIDI scores with corresponding SATB a cappella arrangements, forming the basis for a deep learning system that generates SATB choruses. A subsequent rule-based module then creates a suitable vocal percussion part to complement the music. Overall, this prototype bridges the gap between traditional manual transcription methods and modern AI-driven music processing, paving the way towards automated creation of accessible, high-quality a cappella arrangements.