@misc{indiciae1a36534e59ac,
  title = {Audio Token Attention Is Predictable Before the Language Model Runs},
  author = {Kyoungjun Park and Yunzhe Li and Lili Qiu},
  year = {2026},
  url = {https://arxiv.org/abs/2609.38878},
  note = {Source identifier: 2609.38878}
}