Saved in:
| Main Authors: | , , , , , , |
|---|---|
| Format: | Preprint |
| Published: |
2026
|
| Subjects: | |
| Online Access: | https://arxiv.org/abs/2605.23261 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866917523139592192 |
|---|---|
| author | Wang, Yuanyuan Yang, Dongchao Deng, Yayue Wu, Zhiyong Guo, Yiwen Meng, Helen Wu, Xixin |
| author_facet | Wang, Yuanyuan Yang, Dongchao Deng, Yayue Wu, Zhiyong Guo, Yiwen Meng, Helen Wu, Xixin |
| contents | Evaluating speech generation still relies heavily on human judgments, such as Mean Opinion Score (MOS), which are expensive, subjective, and difficult to reproduce at scale. While a few recent studies have begun to explore AudioLLM-based judge models, existing efforts typically target only a narrow set of scenarios (e.g., utterance-level quality or single-turn dialogue) and provide limited coverage of diverse speech generation tasks and evaluation dimensions. In this work, we propose UniSRM, a unified speech reward model that can support multi-dimensional, interpretable reward signals with reliable reasoning. To support training and evaluation, we introduce UniSRM-Data and UniSRM-Bench, covering speech evaluation tasks from utterance-level quality to context-level coherence. Based on this dataset, we present the unified speech reward model, UniSRM, with a two-stage pipeline that enables reasoning-based fine-grained assessment. Furthermore, we introduce Reasoning-Consistent Rewards to improve the reliability of the reasoning process. Experiments show that UniSRM delivers more reliable and human-aligned judgments across a broad range of speech evaluation tasks, offering a practical foundation for scalable and unified evaluation of speech quality. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2605_23261 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | UniSRM: A Unified Speech Reward Model for Reasoning-Based Fine-grained Assessment Wang, Yuanyuan Yang, Dongchao Deng, Yayue Wu, Zhiyong Guo, Yiwen Meng, Helen Wu, Xixin Audio and Speech Processing Sound Evaluating speech generation still relies heavily on human judgments, such as Mean Opinion Score (MOS), which are expensive, subjective, and difficult to reproduce at scale. While a few recent studies have begun to explore AudioLLM-based judge models, existing efforts typically target only a narrow set of scenarios (e.g., utterance-level quality or single-turn dialogue) and provide limited coverage of diverse speech generation tasks and evaluation dimensions. In this work, we propose UniSRM, a unified speech reward model that can support multi-dimensional, interpretable reward signals with reliable reasoning. To support training and evaluation, we introduce UniSRM-Data and UniSRM-Bench, covering speech evaluation tasks from utterance-level quality to context-level coherence. Based on this dataset, we present the unified speech reward model, UniSRM, with a two-stage pipeline that enables reasoning-based fine-grained assessment. Furthermore, we introduce Reasoning-Consistent Rewards to improve the reliability of the reasoning process. Experiments show that UniSRM delivers more reliable and human-aligned judgments across a broad range of speech evaluation tasks, offering a practical foundation for scalable and unified evaluation of speech quality. |
| title | UniSRM: A Unified Speech Reward Model for Reasoning-Based Fine-grained Assessment |
| topic | Audio and Speech Processing Sound |
| url | https://arxiv.org/abs/2605.23261 |