The authors built AURAL (Adaptive Unspoken Reasoning over Alternative Latents) to address the latency cost of explicit chain-of-thought (CoT) in speech language models.
Latent reasoning cuts speech model response delay by roughly 12x
AURAL models multiple reasoning paths in latent space and jointly predicts future states, matching chain-of-thought quality with far lower latency.
Chinese Tech
Yuxiang Wang · Kunyu Feng · Yuancheng Wang · Zihang Liu · Shengbo Cai · Qinke Ni · +9 more
The Chinese University of Hong Kong, Shenzhen · Tencent Hunyuan · Tsinghua University · The Hong Kong University of Science and Technology · Amphion Technology Co., Ltd.
Research Digest··2 min read
The authors introduce AURAL, a latent reasoning framework for speech language models that generates plausible reasoning continuations in continuous hidden states instead of text tokens.
Why this paper
From Tencent Hunyuan and 4 others
In one line
A latent reasoning method that models multiple reasoning paths and predicts future states in chunks rivals explicit chain-of-thought in speech models while cutting time to first answer by 11.8x.
What we could check
- ·No code link found
- ·No weights link found
- ·No dataset link found
- ·No compute details found
- ·No stated limitations found
- ·No benchmark numbers found
Observed from the paper text and links we have. Absence here means we did not find it, not that it does not exist.
§