Text-to-music systems produce increasingly convincing audio, yet evaluation reveals little about whether the result matches user intent. A global text-audio …
机构:上交
来源:arXiv 2610.10355 | AI4Papers 论文推荐平台