TY  - RPRT
TI  - Batched Speech Decisions Without Decoding: Single-Token Supervision Lets a Frozen LLM Hear Beyond the Transcript
AU  - Jie Jin
AU  - Ziyin Ma
AU  - Min Yin
AU  - Jinyu Chen
AU  - Haigang Song
AU  - Zhikun Pang
AU  - Xiaowen Zhang
PY  - 2026
UR  - https://arxiv.org/abs/2610.02638
ID  - 2610.02638
ER  -