Emotionally consistent music melody generation algorithm integrating prompt perception and hyper-network optimization

Area of Science:

• Artificial Intelligence
• Music Technology
• Computational Musicology

Background:

• Intelligent music melody generation faces challenges in maintaining emotion consistency, understanding user intent, and adapting dynamically.
• Existing algorithms often struggle with integrating diverse cues and adhering to music theory constraints.

Purpose of the Study:

• To propose a stable emotion-consistent music melody generation algorithm (ECM-HPO) that addresses limitations in current intelligent melody generation.
• To enhance emotion consistency, user intent interpretation, and dynamic adaptability in generated melodies.

Main Methods:

• Developed an emotion consistency enhancement mechanism using a unified emotion space combining audio, text, and historical melody data.
• Implemented a cue-aware melody-guided generation module with a unified encoding framework and multi-scale cross-attention for semantic alignment.
• Designed a hyper-network-optimized dynamic generation architecture with a three-level hyper-network and a music-specific search space incorporating music theory constraints.

Main Results:

• The ECM-HPO algorithm demonstrated superior performance over baseline methods on the EMD-Melody dataset.
• Achieved high scores in melody contour smoothness (PCS: 0.910), rhythm consistency (RC: 0.891), and emotion recognition accuracy (ERA: 92.4%).
• Ablation studies confirmed the effectiveness of individual modules, with the complete model showing a 15.19% improvement in PCS over the basic model.

Conclusions:

• The proposed ECM-HPO algorithm offers a robust and effective solution for emotion-consistent music melody generation.
• The integration of cue-awareness and hyper-network optimization significantly improves melody quality and adaptability.
• The algorithm's cross-style robustness suggests broad applicability in music generation tasks.