A melody perception music re-orchestration method and system based on structure-texture feature decoupling
By constructing a weakly paired dataset and training a contrastive melody encoder, acoustic texture information is extracted, and dual-flow control is combined to solve the problems of feature coupling and text control failure in melody conditional music generation, thus achieving high-fidelity melody adaptation across styles and instruments.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- SHANGHAI ARTIFICIAL INTELLIGENCE INNOVATION CENT
- Filing Date
- 2026-04-22
- Publication Date
- 2026-07-10
AI Technical Summary
In existing melody conditional music generation methods, the coupling of structural and texture features leads to model overfitting and text control failure, making it impossible to effectively achieve melody adaptation across styles and instruments.
By constructing a weakly paired dataset, a contrastive melody encoder and an adaptation generation model are trained. Contrastive learning and sparsification are used to strip away acoustic texture information, extract pure melody structure representations, and combine dual flow conditional control to generate music that conforms to the target audio information.
It enables cross-style and cross-instrument melody adaptation, improves text control compliance and model generalization ability, ensures melody consistency and audio quality, and adapts to music generation in complex scenarios.
Smart Images

Figure CN122073114B_ABST
Abstract
Citation Information
Patent Citations
Classroom multi-modal data processing method and system based on deep learning
CN120850228A
Simultaneous interpretation data processing method and system based on POE microphone array
CN121354578A