REFFLY: Melody-Constrained Lyrics Editing Model
Songyan Zhao, Bingxuan Li, Yufei Tian, and Nanyun Peng, in Proceedings of the 2025 Annual Conference of the North American Chapter of the Association for Computational Linguistics (NAACL), 2025.
Download the full text
Abstract
Automatic melody-to-lyric (M2L) generation seeks lyrics that fit a given tune. Prior systems generate from scratch, offering limited control and poor alignment. We present REFFLY—the first \emphrevision framework that edits arbitrary plain-text drafts into full, melody-aligned lyrics. Trained on a synthesized melody-aligned lyrics dataset and enhanced with training-free semantics- and musicality-preserving heuristics, REFFLY handles tasks such as flexible lyric drafting, song translation, and style transfer. Experiments show it outperforms strong baselines (Lyra and GPT-4) by 25% in both musicality and text quality.
Bib Entry
@inproceedings{zhao2025reffly, author = {Zhao, Songyan and Li, Bingxuan and Tian, Yufei and Peng, Nanyun}, title = {REFFLY: Melody-Constrained Lyrics Editing Model}, booktitle = {Proceedings of the 2025 Annual Conference of the North American Chapter of the Association for Computational Linguistics (NAACL)}, year = {2025} }