Audio language models are meant to answer questions from the audio itself, but an hour-long recording rarely fits in their context, and the answer often depe…
机构:CUHK
来源:arXiv 2611.04480 | AI4Papers 论文推荐平台