Paper Detail
Yuxin Zhang, Xiangyu Tony Zhang, Daijiao Liu, Fei Tian, Yayue Deng, Jun Chen, Qingjian Lin, Haoyang Zhang, Yuxin Li, Jinglan Gong, Yechang Huang, Liang Zhao, Chengyuan Yao, Hexin Liu, Eng Siong Chng, Xuerui Yang, Gang Yu, Xiangyu Zhang, Daxin Jiang
Recent advancements in large audio language models have extended Chain-of-Thought (CoT) reasoning into the auditory domain, enabling models to tackle increasingly complex acoustic and spoken tasks. To elicit and sustain these extended reasoning chains, the prevailing paradigm -- driven by the success of text-based reasoning models -- overwhelmingly relies on Reinforcement Learning with Verified Rewards (RLVR). However, as models are strictly optimized to distill rich, continuous auditory contexts into isolated, verifiable text labels, a fundamental question arises: are we fostering true audio intelligence, or merely reducing a continuous sensory medium into a discrete puzzle? We identify this as the "verifiable reward trap." While RLVR yields remarkable scores on standardized objective benchmarks, it systematically degrades the real-world conversational feel of audio models. By prioritizing isolated correctness over acoustic nuance, RLVR reduces dynamic interactions to mechanical "answering machines," severely compromising prosodic naturalness, emotional continuity, and user immersion, particularly in long-turn dialogues. To bridge the gap between mechanical objective verification and genuine sensory empathy, we introduce Step-Audio-R1.5, marking a paradigm shift toward Reinforcement Learning from Human Feedback (RLHF) in audio reasoning. Comprehensive evaluations demonstrate that Step-Audio-R1.5 not only maintains robust analytical reasoning but profoundly transforms the interactive experience, redefining the boundaries of deeply immersive long-turn spoken dialogue.
No structured notes yet. Add `summary_sections`, `why_relevant`, `claim_impact`, or `next_action` in `papers.jsonl` to enrich this view.
No ranking explanation is available yet.
No tags.
@misc{zhang2026step,
title = {Step-Audio-R1.5 Technical Report},
author = {Yuxin Zhang and Xiangyu Tony Zhang and Daijiao Liu and Fei Tian and Yayue Deng and Jun Chen and Qingjian Lin and Haoyang Zhang and Yuxin Li and Jinglan Gong and Yechang Huang and Liang Zhao and Chengyuan Yao and Hexin Liu and Eng Siong Chng and Xuerui Yang and Gang Yu and Xiangyu Zhang and Daxin Jiang},
year = {2026},
abstract = {Recent advancements in large audio language models have extended Chain-of-Thought (CoT) reasoning into the auditory domain, enabling models to tackle increasingly complex acoustic and spoken tasks. To elicit and sustain these extended reasoning chains, the prevailing paradigm -- driven by the success of text-based reasoning models -- overwhelmingly relies on Reinforcement Learning with Verified Rewards (RLVR). However, as models are strictly optimized to distill rich, continuous auditory context},
url = {https://huggingface.co/papers/2604.25719},
keywords = {Chain-of-Thought, Reinforcement Learning with Verified Rewards, RLVR, Reinforcement Learning from Human Feedback, RLHF, audio language models, auditory domain, acoustic contexts, verifiable reward trap, long-turn dialogues, prosodic naturalness, emotional continuity, user immersion, code available, huggingface daily},
eprint = {2604.25719},
archiveprefix = {arXiv},
}
{}