AI
YuE2: Unifying Symbolic and Audio Music Generation at Frontier Quality
Publication: Hugging Face Daily Papers
Author: Not available in collected metadata
Source published: 26/09/2026, 20:00:00 UTC
Briefing created: 29/09/2026, 02:29:00 UTC
Based on a feed excerpt; full article content has not been acquired. Source evidence retrieved 29/09/2026, 02:18:24 UTC.
Briefings
YuE2 unifies symbolic and audio music generation at frontier quality using a single AR-NAR MoT that writes explicit scores, expands them to semantic tokens, and realizes as full-song audio. On WildSongBench, it scored 6.73 on SongBench Global Avg, beating all evaluated baselines. Best-of-8 selection reached 6.96, the highest mean among systems. Experts prefer YuE2 over separate language models and diffusion Transformers, and favor best-of-8 over Suno v4.5 and Suno v5. To enable this, MERT2 and SheetSage2 provide semantic and symbolic supervision, achieving state-of-the-art performance on MARBLE metrics and lead-sheet transcription benchmarks.
Key points
- YuE2 uses a single AR-NAR Mixture-of-Transformers (MoT) to write explicit scores, expand them into semantic music tokens, and realize them as full-song audio.
- On WildSongBench, YuE2 scored 6.73 on SongBench Global Avg, exceeding all evaluated public baselines and achieving the highest mean score among all evaluated systems.
Context and limitations
Coverage is limited to WildSongBench and SongBench Global Avg metrics; results are based on expert preference surveys and best-of-8 selections rather than full-scale production testing.
Generated by Bittrees-hosted qwen35-2b sha256:aaf42c8b7c3c on 29/09/2026, 02:29:00 UTC. Based on publisher-provided feed evidence; may not cover the full original article. Check the source for details.
View archived briefing on IPFS
bafybeidr4blgoaqt7pbajg7mhtjoamzyvn2grgqldkjlaltmudts4x3dui