Why This Matters
YuE2 is a music generation system that pairs symbolic planning with audio generation, evaluated on WildSongBench, a benchmark of 192 prompts across 15 system settings. The reported results emphasize song quality, text alignment, and production quality, plus a best-of-8 sampling strategy that picks among eight generations by musicality, prompt control, and lyric accuracy. It signals continued push toward controllable, lyric-accurate AI music that could reshape production tooling and rights debates.
Key Takeaways
- WildSongBench evaluates 192 prompts across 15 system settings using automatic metrics.
- Best-of-8 sampling selects generations by musicality, prompt control, and lyric accuracy.
- Scoring blends SongBench/SongEval for quality and MuLan/AllMusicCaps/prompt control for text alignment, with AudioBox production quality shown as bubble area.
WildSongBench. 192 prompts and 15 system settings. The table reports automatic evaluation scores. Best-of-8 selects one of eight generations by musicality, prompt control, and lyric accuracy.
Figure 1. Song quality combines SongBench and SongEval; text alignment combines MuLan, AllMusicCaps, and prompt control. Both axes show normalized comparison indices. Bubble area represents AudioBox production quality.