Elamur Research⌕
☰
home / papers / 2025 / SoundMind: RL-Incentivized Logic Reasoning for Audio-Language Models

SoundMind: RL-Incentivized Logic Reasoning for Audio-Language Models

Paper · 2025 · Xingjian Diao, Chunhui Zhang, Keyi Kong et al.

Summary

Large audio-language models (LALMs) can understand speech, but they are weak at multi-step logical reasoning. A key reason is that existing audio reasoning corpora mostly give short text labels without spoken reasoning traces. The authors build SoundMind, a speech benchmark for audio logical reasoning (ALR), derived from LogiQA 2.0–NLI. Each premise–conclusion triplet is first rewritten into a conversational prompt. An LLM then writes a chain-of-thought explanation and an "entailed"/"not-entailed" verdict. Finally, a TTS model voices both the prompt and the response, so text and audio stay aligned.

They also propose SoundMind-RL, a rule-based reinforcement learning recipe. It builds on REINFORCE++, borrows reward ideas from DeepSeek-R1, and adds a strict format reward inspired by Logic-RL to discourage text-only shortcuts. Fine-tuning Qwen2.5-Omni-7B this way raises accuracy across text-to-audio, audio-to-text and audio-to-audio settings, outperforming the baselines tested. The trade-off is higher WER in the generated speech.

The work matters because it supplies both a resource and a training method for reasoning over speech, which earlier work had largely left unexplored.

Key points

  • SoundMind dataset: 6,446 samples with over 1,074 hours of speech. Each sample includes user content, CoT reasoning, the answer, and input/output audio. The speech is synthetic and the CoT is LLM-generated. Code and data are at github.com/xid32/SoundMind.
  • Text-to-audio results: Qwen2.5-Omni-7B accuracy rises from 80.79% to 83.84% with SoundMind-RL. WER goes from 2.18% to 6.99%.
  • Audio-to-audio results: accuracy improves from 77.59% to 81.40%, while WER increases from 2.23% to 8.95%. The authors attribute the WER increase to longer reasoning outputs and the lack of a fluency reward.
  • Audio-to-text baselines: MiniCPM-o, Gemini-Pro-V1.5, Baichuan-Omni-1.5 and Qwen2-Audio were compared against the base and RL-tuned Qwen2.5-Omni-7B.
  • Limitations noted by the authors: the rule-based rewards may be too rigid, synthetic data may introduce artifacts, and there appears to be a trade-off between reasoning depth and speech fluency.

Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.

 Institutions

🎓Dartmouth CollegeHanover · United States37 works · 42,625 citations🎓Shandong UniversityJinan · China17 works · 789 citations

 Fields, topics and concepts

📚 Shelves 6The field of science this work belongs to (OpenAlex fields).Computer Science · 16,667 worksArtificial Intelligence · 16,194 worksArts and Humanities · 16,039 worksLanguage and Linguistics · 15,891 worksComputer Vision and Pattern Recognition · 2,869 worksSignal Processing · 452 works

🧭 Topics 3Research subjects and languages the work is about.Language models & pretraining · 3,320 worksReasoning · 1,426 worksSpeech & spoken language · 480 works

Cite

Cite as Xingjian Diao, Chunhui Zhang, Keyi Kong, Weiyi Wu, Chiyu Ma, Zhongyu Ouyang, Peijun Qing, Soroush Vosoughi and Jiang Gui. 2025. SoundMind: RL-Incentivized Logic Reasoning for Audio-Language Models. Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing.

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.18653/v1/2025.emnlp-main.27

@inproceedings{diao2025soundmind,
  title = {{SoundMind: RL-Incentivized Logic Reasoning for Audio-Language Models}},
  author = {Xingjian Diao, Chunhui Zhang, Keyi Kong, Weiyi Wu, Chiyu Ma, Zhongyu Ouyang, Peijun Qing, Soroush Vosoughi and Jiang Gui},
  booktitle = {Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing},
  year = {2025},
  doi = {10.18653/v1/2025.emnlp-main.27},
  url = {https://doi.org/10.18653/v1/2025.emnlp-main.27}
}

References

47 references, 18 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Aleksander Madry, Alex Beutel, Alex Carney, et al. 2024. Openai o1 system card. arXiv preprint arXiv:2412.16720.
  2. Martijn Bartelds, Nay San, Bradley McDonnell, Dan Jurafsky, and Martijn Wieling. 2023. Making more of little data: Improving low-resource automatic speech recognition using data augmentation. arXiv preprint arXiv:2305.10951. find
  3. Ye Jia, Yifan Ding, Ankur Bapna, Colin Cherry, Yu Zhang, Alexis Conneau, and Nobuyuki Morioka. 2022. Leveraging unsupervised and weakly-supervised data to improve direct speech-to-speech translation. arXiv preprint arXiv:2203.13339. find
  4. Jean Carletta, Simone Ashby, Sebastien Bourban, Mike Flynn, Mael Guillemot, Thomas Hain, Jaroslav Kadlec, Vasilis Karaiskos, Wessel Kraaij, Melissa Kronenthal, et al. 2005. The ami meeting corpus: A pre-announcement. In International Workshop on Machine Learning for Multimodal Interaction. find
  5. Ming Cheng, Xingjian Diao, Shitong Cheng, and Wenjun Liu. 2024. Saic: Integration of speech anonymization and identity classification. In AI for Health Equity and Fairness: Leveraging AI to Address Social Determinants of Health, pages 295–306. Springer. find
  6. Yunfei Chu, Jin Xu, Qian Yang, Haojie Wei, Xipin Wei, Zhifang Guo, Yichong Leng, Yuanjun Lv, Jinzheng He, Junyang Lin, et al. 2024. Qwen2-audio technical report. arXiv preprint arXiv:2407.10759.
  7. Xingjian Diao, Chunhui Zhang, Tingxuan Wu, Ming Cheng, Zhongyu Ouyang, Weiyi Wu, and Jiang Gui. 2025. Learning musical representations for music performance question answering. arXiv preprint arXiv:2502.06710. find
  8. Sreyan Ghosh, Zhifeng Kong, Sonal Kumar, S Sakshi, Jaehyeon Kim, Wei Ping, Rafael Valle, Dinesh Manocha, and Bryan Catanzaro. 2025. Audio flamingo 2: An audio-language model with long-audio understanding and expert reasoning abilities. arXiv preprint arXiv:2503.03983. — [arXiv:2503.03983](https://arxiv.org/abs/2503.03983) find
  9. Daya Guo, Dejian Yang, Haowei Zhang, Junxiao Song, Ruoyu Zhang, Runxin Xu, Qihao Zhu, Shirong Ma, Peiyi Wang, Xiao Bi, et al. 2025. Deepseek-r1: Incentivizing reasoning capability in llms via reinforcement learning. arXiv preprint arXiv:2501.12948.
  10. Jarvis Guo, Tuney Zheng, Yuelin Bai, Bo Li, Yubo Wang, King Zhu, Yizhi Li, Graham Neubig, Wenhu Chen, and Xiang Yue. 2024. Mammoth-vl: Eliciting multimodal reasoning with instruction tuning at scale. arXiv preprint arXiv:2412.05237.
  11. Ziyue Jiang, Yi Ren, Ruiqi Li, Shengpeng Ji, Boyang Zhang, Zhenhui Ye, Chen Zhang, Bai Jionghao, Xiaoda Yang, Jialong Zuo, et al. 2025. Megatts 3: Sparse alignment enhanced latent diffusion transformer for zero-shot speech synthesis. arXiv preprint arXiv:2502.18924. — [arXiv:2502.18924](https://arxiv.org/abs/2502.18924) find
  12. Feihu Jin, Yifan Liu, and Ying Tan. 2024. Zero-shot chain-of-thought reasoning guided by evolutionary algorithms in large language models. arXiv preprint arXiv:2402.05376.
  13. Kimi Team, Angang Du, Bofei Gao, Bowei Xing, Changjiu Jiang, Cheng Chen, Cheng Li, Chenjun Xiao, Chenzhuang Du, Chonghua Liao, et al. 2025. Kimi k1. 5: Scaling reinforcement learning with llms. arXiv preprint arXiv:2501.12599. — Kimi k1.5: Scaling Reinforcement Learning with LLMs
  14. Zhifeng Kong, Arushi Goel, Rohan Badlani, Wei Ping, Rafael Valle, and Bryan Catanzaro. 2024. Audio flamingo: A novel audio language model with few-shot learning and dialogue abilities. In International Conference on Machine Learning. find
  15. Bin Li, Bin Sun, Shutao Li, Encheng Chen, Hongru Liu, Yixuan Weng, Yongping Bai, and Meiling Hu. 2024. Distinct but correct: generating diversified and entity-revised medical response. Science China Information Sciences. find
  16. Yadong Li, Jun Liu, Tao Zhang, Song Chen, Tianpeng Li, Zehuan Li, Lijun Liu, Lingfeng Ming, Guosheng Dong, Da Pan, et al. 2025. Baichuan-omni-1.5 technical report. arXiv preprint arXiv:2501.15368.
  17. Hunter Lightman, Vineet Kosaraju, Yuri Burda, Harrison Edwards, Bowen Baker, Teddy Lee, Jan Leike, John Schulman, Ilya Sutskever, and Karl Cobbe. 2023. Let’s verify step by step. In International Conference on Learning Representations.
  18. Xiaotian Han, Yiren Jian, Xuefeng Hu, Haogeng Liu, Yiqi Wang, Qihang Fan, Yuang Ai, Huaibo Huang, Ran He, Zhenheng Yang, et al. 2024. Infimm-webmath-40b: Advancing multimodal pre-training for enhanced mathematical reasoning. arXiv preprint arXiv:2409.12568. find
  19. Hanmeng Liu, Jian Liu, Leyang Cui, Zhiyang Teng, Nan Duan, Ming Zhou, and Yue Zhang. 2023. Logiqa 2.0—an improved dataset for logical reasoning in natural language understanding. Transactions on Audio, Speech, and Language Processing. find
  20. Jian Hu, Jason Klein Liu, Haotian Xu, and Wei Shen. 2025. Reinforce++: An efficient rlhf algorithm with robustness to both prompt and reward models. arXiv preprint arXiv:2501.03262. find
  21. Ziyang Ma, Zhuo Chen, Yuping Wang, Eng Siong Chng, and Xie Chen. 2025. Audio-cot: Exploring chain-of-thought reasoning in large audio language model. arXiv preprint arXiv:2501.07246. — [arXiv:2501.07246](https://arxiv.org/abs/2501.07246) find
  22. Aaron Jaech, Adam Kalai, Adam Lerer, Adam Richardson, Ahmed El-Kishky, Aiden Low, Alec Helyar, OpenAI. 2024. Whisperspeech. https://github.com/WhisperSpeech/WhisperSpeech. ↗
  23. Paribesh Regmi, Arjun Dahal, and Basanta Joshi. 2019. Nepali speech recognition using rnn-ctc model. International Journal of Computer Applications. find
  24. Hao Shao, Shengju Qian, Han Xiao, Guanglu Song, Zhuofan Zong, Letian Wang, Yu Liu, and Hongsheng Li. 2024. Visual cot: Advancing multi-modal language models with a comprehensive dataset and benchmark for chain-of-thought reasoning. In The Thirty-eight Conference on Neural Information Processing Syst… find
  25. Zayne Rea Sprague, Fangcong Yin, Juan Diego Rodriguez, Dongwei Jiang, Manya Wadhwa, Prasann Singhal, Xinyu Zhao, Xi Ye, Kyle Mahowald, and Greg Durrett. 2025. To cot or not to cot? chain-of-thought helps mainly on math and symbolic reasoning. In The Thirteenth International Conference on Learning Re…
  26. Mirac Suzgun, Nathan Scales, Nathanael Schärli, Sebastian Gehrmann, Yi Tay, Hyung Won Chung, Aakanksha Chowdhery, Quoc Le, Ed Chi, Denny Zhou, et al. 2023. Challenging big-bench tasks and whether chain-of-thought can solve them. In Findings of the Association for Computational Linguistics: ACL.
  27. Gemini Team, Petko Georgiev, Ving Ian Lei, Ryan Burnell, Libin Bai, Anmol Gulati, Garrett Tanzer, Damien Vincent, Zhufeng Pan, Shibo Wang, et al. 2024. Gemini 1.5: Unlocking multimodal understanding across millions of tokens of context. arXiv preprint arXiv:2403.05530.
  28. Ziyu Wan, Xidong Feng, Muning Wen, Stephen Marcus McAleer, Ying Wen, Weinan Zhang, and Jun Wang. 2024. Alphazero-like tree-search can guide large language model decoding and training. In International Conference on Machine Learning. find
  29. Wengran Wang, Yudong Rao, Rui Zhi, Samiha Marwan, Ge Gao, and Thomas W Price. 2020. Step tutor: Supporting students through step-by-step example-based feedback. In ACM conference on Innovation and Technology in Computer Science Education. find
  30. Xinsheng Wang, Mingqi Jiang, Ziyang Ma, Ziyu Zhang, Songxiang Liu, Linqin Li, Zheng Liang, Qixi Zheng, Rui Wang, Xiaoqin Feng, et al. 2025. Spark-tts: An efficient llm-based text-to-speech model with single-stream decoupled speech tokens. arXiv preprint arXiv:2503.01710. — [arXiv:2503.01710](https://arxiv.org/abs/2503.01710) find
  31. Jason Wei, Xuezhi Wang, Dale Schuurmans, Maarten Bosma, Fei Xia, Ed Chi, Quoc V Le, Denny Zhou, et al. 2022. Chain-of-thought prompting elicits reasoning in large language models. In Advances in neural information processing systems.
  32. Tian Xie, Zitian Gao, Qingnan Ren, Haoming Luo, Yuqian Hong, Bryan Dai, Joey Zhou, Kai Qiu, Zhirong Wu, and Chong Luo. 2025a. Logic-rl: Unleashing llm reasoning with rule-based reinforcement learning. arXiv preprint arXiv:2502.14768.
  33. Zhifei Xie, Mingbao Lin, Zihang Liu, Pengcheng Wu, Shuicheng Yan, and Chunyan Miao. 2025b. Audio-reasoner: Improving reasoning capability in large audio language models. arXiv preprint arXiv:2503.02318.
  34. Guowei Xu, Peng Jin, Li Hao, Yibing Song, Lichao Sun, and Li Yuan. 2024. Llava-o1: Let vision language models reason step-by-step. arXiv preprint arXiv:2411.10440. find
  35. Jin Xu, Zhifang Guo, Jinzheng He, Hangrui Hu, Ting He, Shuai Bai, Keqin Chen, Jialin Wang, Yang Fan, Kai Dang, et al. 2025. Qwen2. 5-omni technical report. arXiv preprint arXiv:2503.20215. — Qwen2.5-Omni Technical Report
  36. Shunyu Yao, Dian Yu, Jeffrey Zhao, Izhak Shafran, Tom Griffiths, Yuan Cao, and Karthik Narasimhan. 2023. Tree of thoughts: Deliberate problem solving with large language models. In Advances in neural information processing systems. find
  37. Yuan Yao, Tianyu Yu, Ao Zhang, Chongyi Wang, Junbo Cui, Hongji Zhu, Tianchi Cai, Haoyu Li, Weilin Zhao, Zhihui He, et al. 2024. Minicpm-v: A gpt-4v level mllm on your phone. arXiv preprint arXiv:2408.01800.
  38. Wenhao You, Xingjian Diao, Chunhui Zhang, Keyi Kong, Weiyi Wu, Zhongyu Ouyang, Chiyu Ma, Tingxuan Wu, Noah Wei, Zong Ke, et al. 2025. Music’s multimodal complexity in avqa: Why we need more than general multimodal llms. arXiv preprint arXiv:2505.20638. — [arXiv:2505.20638](https://arxiv.org/abs/2505.20638) find
  39. Xiangchi Yuan, Chunhui Zhang, Zheyuan Liu, Dachuan Shi, Leyan Pan, Soroush Vosoughi, and Wenke Lee. 2025. Superficial self-improved reasoners benefit from model merging. In Proceedings of the Conference on Empirical Methods in Natural Language Processing.
  40. Rodolfo Zevallos. 2022. Text-to-speech data augmentation for low resource speech recognition. arXiv preprint arXiv:2204.00291. — [arXiv:2204.00291](https://arxiv.org/abs/2204.00291) find
  41. Xiao Zhan, Noura Abdi, William Seymour, and Jose Such. 2024. Healthcare voice ai assistants: factors influencing trust and intention to use. ACM on Human-Computer Interaction. find
  42. Chunhui Zhang, Yiren Jian, Zhongyu Ouyang, and Soroush Vosoughi. 2024. Working memory identifies reasoning limits in language models. In Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing. find
  43. Chunhui Zhang, Zhongyu Ouyang, Kwonjoon Lee, Nakul Agarwal, Sean Dae Houlihan, Soroush Vosoughi, and Shao-Yuan Lo. 2025a. Overcoming multi-step complexity in multimodal theory-of-mind reasoning: A scalable bayesian planner. In International Conference on Machine Learning. find
  44. Chunhui Zhang, Sirui Wang, Zhongyu Ouyang, Xiangchi Yuan, and Soroush Vosoughi. 2025b. Growing through experience: Scaling episodic grounding in language models. In Annual Meeting of the Association for Computational Linguistics.
  45. Yu Zhao, Huifeng Yin, Bo Zeng, Hao Wang, Tianqi Shi, Chenyang Lyu, Longyue Wang, Weihua Luo, and Kaifu Zhang. 2024. Marco-o1: Towards open reasoning models for open-ended solutions. arXiv preprint arXiv:2411.14405. find
  46. Guolong Zhong, Hongyu Song, Ruoyu Wang, Lei Sun, Diyuan Liu, Jia Pan, Xin Fang, Jun Du, J. Zhang, and Lirong Dai. 2022. External text based data augmentation for low-resource speech recognition in the constrained condition of openasr21 challenge. In Interspeech. find
  47. Zyphra. 2024. Zonos. https://github.com/Zyphra/ Zonos. ↗

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/diao-2025-soundmind-rl-incentivized-logic-reasoning/