titleDecode Before You Hear: Speculative Listening for Chunk-Interleaved Full-Duplex Speech Language Models
authorsYutong Bian, Yanglin Zhang, Yuxiang Wang, Luoyuan Zhang, Rongshen He, Zihao Wang, Junwen Qiu, Jiaqi Li, Zhizheng Wu
date2026-11-04
venuearXiv (preprint, not peer reviewed)
idarXiv:2611.06982
affiliationChinese University of Hong Kong, Shenzhen
upvotes5 (Hugging Face Daily Papers, 2026-11-08)
linksabs pdf hf
cite
relatedSolarWM: Open Data and Scalable Training for Long-Horizon Video World Models
GameCraft-Bench: Can Agents Build Playable Games End-to-End in a Real Game Engine?
TRL-Bench: Standardizing Cross-Paradigm Representation-Level Evaluation of Tabular Encoders
R^2Orch: Recursive Self-Improvement for Agent Orchestration via Self-Play within a Skill-Internalization Loop
COBRA-Skills: Contextual Bandit-Guided Evolution for Agent Skill Optimization
abstractChunk-interleaved full-duplex speech language models listen and speak in fixed time chunks, and each chunk must be processed before the next one arrives. Their workload is asymmetric across chunks: listen chunks leave compute idle while awaiting audio, and speak and tool-call chunks require serial generation, which delays responses and tool calls. Using the idle compute for the next chunk requires decoding it before its audio arrives, under an assumed input. Taking silence as the assumed input and comparing the output with the output under real audio on four models, we observe two properties. First, the next audio chunk does not change the output in 95.3% of chunks. Second, 88% of changes begin at the first output token. Building on these properties, we propose Speculative Listening, a training-free, lossless framework for input-time self-speculative decoding. During idle time, the model drafts the next chunk under silence and, for changes at the first token, a second branch from the runner-up first token. It keeps all drafts in a cross-chunk hypothesis pool and verifies them when the audio arrives. On our eight-task benchmark TickTock-Bench, Speculative Listening speeds up speak chunks by 2.6-2.8x and tool-call chunks by up to 23.4x. On average, it reduces the decoding delay before a response starts by 73%, to 74-241 ms, close to the 200 ms gap between human turns, and the maximum lag behind real time by 94.4%.

bibtex

@misc{bian2026decode,
  title={Decode Before You Hear: Speculative Listening for Chunk-Interleaved Full-Duplex Speech Language Models},
  author={Yutong Bian and Yanglin Zhang and Yuxiang Wang and Luoyuan Zhang and Rongshen He and Zihao Wang and Junwen Qiu and Jiaqi Li and Zhizheng Wu},
  year={2026},
  eprint={2611.06982},
  archivePrefix={arXiv},
  url={https://arxiv.org/abs/2611.06982}
}