Segment-Factorized Full-Song Generation on Symbolic Piano Music

2025-10-07

AI Summary

This study addresses the challenge of jointly ensuring structural controllability and musical coherence in fully automated symbolic piano composition. To this end, we propose the Segmented Full-Score (SFS) model, which decomposes a composition into user-defined structural segments (e.g., verse, chorus) and models each segment separately. SFS introduces segmented factorized representations and selective cross-segment attention to preserve long-term structural consistency while enhancing local melodic fluency. Built upon the Transformer architecture, the model is integrated into a web-based piano roll interface, supporting seed-guided generation, customizable structural templates, and real-time interactive editing. Quantitative and qualitative evaluations demonstrate that SFS significantly outperforms existing methods in structural accuracy and melodic coherence for complete piece generation. The system has been deployed as an online human-AI collaborative composition platform.

Technology Category

• Natural Language Processing: Code Generation / Program Synthesis from Natural Language
• Humans and AI: Human-Aware Planning and Behavior Prediction
• Machine Learning: Neuro-Symbolic Learning

Application Category

• Semantics and Knowledge: Methods, algorithms and applications for the development of semantic models, knowledge graphs and other forms of structured data models with machine-interpretable semantics
• Responsible Web: Machine-in-the-loop, human agency and autonomy
• Economics, Online Markets and Human Computation: Economic ramifications for generative AI infrastructure and applications

Abstract

We propose the Segmented Full-Song Model (SFS) for symbolic full-song generation. The model accepts a user-provided song structure and an optional short seed segment that anchors the main idea around which the song is developed. By factorizing a song into segments and generating each one through selective attention to related segments, the model achieves higher quality and efficiency compared to prior work. To demonstrate its suitability for human-AI interaction, we further wrap SFS into a web application that enables users to iteratively co-create music on a piano roll with customizable structures and flexible ordering.

Problem

• Generating symbolic full-song piano music
• Factorizing songs into segments for efficient generation
• Enabling human-AI co-creation with customizable structures

Innovation

• Segments generated via selective attention mechanisms
• User-defined song structure with optional seed segments
• Web app enables interactive co-creation on piano roll