YuE2 is a single model that first generates a readable musical score specifying melody and harmony, then expands it into semantic tokens, and finally produces a full song audio recording. In expert listening tests, the version with symbolic planning was preferred 49.3% of the time over 34.6% without planning. The model also outperformed separate language model and diffusion Transformer systems, and scored 6.73 on SongBench Global Avg, exceeding all public baselines. Best-of-8 selection reached 6.96, the highest mean among all evaluated systems, and expert listeners found it competitive with proprietary systems like Suno v4.5 and v5.
YuE2 unifies symbolic and audio music generation via explicit score planning
A single mixture-of-transformers model writes readable scores, expands them into semantic tokens, and generates full-song audio, outperforming prior baselines and rivaling proprietary systems.
Top University
Ruibin Yuan · Jiahao Pan · Junyan Jiang · Zhiyue Wu · Ziya Zhou · Jiankai Sun · +29 more
HKUST · M-A-P · Tokenwave.AI · New York University · Stanford University
Research Digest··2 min read
In one line
YuE2 unifies symbolic and audio music generation by first writing a readable score then producing full song audio.
Before you read this
Two ideas this paper builds onmixture of expertsA neural architecture with many specialized sub-networks, only a few activated per input via a router, balancing capacity and efficiency.88 papers →flow matchingA generative training method that learns an ordinary differential equation mapping noise to data by matching velocity fields along probability paths.173 papers →
Read it as
YuE2 unifies symbolic and audio music generation by first writing an explicit score, then expanding it into semantic tokens and synthesizing full-song audio.
The authors introduce YuE2, a unified model that combines symbolic and audio music generation.
How it works
stage by stageThe Daily Edition
The front page in your inbox each weekday morning: the lead, the picks, and why each one earned its place.