TY  - RPRT
TI  - Audio Token Attention Is Predictable Before the Language Model Runs
AU  - Kyoungjun Park
AU  - Yunzhe Li
AU  - Lili Qiu
PY  - 2026
UR  - https://arxiv.org/abs/2609.38878
ID  - 2609.38878
ER  -