Paper introduces SketchSong for hierarchical song generation

Arxiv CS.SD (Sound/Audio ML)

SketchSong generates complete songs via a two-stage approach: first planning the song structure (sketch), then producing multi-track audio. This improves arrangement coherence compared to end-to-end models.

Arxiv CS.SD (Sound/Audio ML)·Xiaoyue DuanNanxing HuYutang FengXudong YanJiatao ChenJinchao ZhangJie Zhou

8 days ago