Aligning Language Models for Lyric-to-Melody Generation with Rule-Based Musical Constraints
arXiv:2604.18489v1 Announce Type: cross
Abstract: Large Language Models (LLMs) show promise in lyric-to-melody generation, but models trained with Supervised Fine-Tuning (SFT) often produce musically implausible melodies with issues like poor rhythm a…