arXiv cs.CL· Shashi Kumar, Yanis Labrak, Sergio Burdisso, Esa\'u Villatoro-Tello, Petr Motlicek, Kadri Hacioglu, Andreas Stolcke·· 12 小时前AI 评分29
把伪标签当奖励而非目标：用于半监督 ASR 的组相对策略优化（GRPO）
Pseudo-Labels as Rewards, Not Targets: Group-Relative Policy Optimization for Semi-Supervised ASR
AI 导读
研究提出用 Group Relative Policy Optimization（GRPO）把伪标签当作奖励参考而非监督目标，在银行领域基于 Whisper 生成标签时相比 SFT 取得 29.3% 至 56.2% 的相对 WER 改善，且该优势在四种 Whisper 规模、四个教师模型族、三个领域和四种语言上均成立。
正文 · 原文
Abstract:Adapting ASR to a domain with only untranscribed audio typically relies on pseudo-labeling, a strong teacher transcribes the audio and a student is fine-tuned on it. Because supervised fine-tuning (SFT) treats every pseudo-label token as a target, the student inherits the teacher's errors. We show that using the same pseudo-labels as a reward reference for Group Relative Policy Optimization (GRPO) is more robust to label noise. In the banking domain, GRPO yields 29.3% to 56.2% relative WER improvements over SFT on Whisper-generated labels, and the advantage holds across four Whisper sizes, four teacher families, three domains and four languages. While SFT reproduces almost 100% of teacher errors verbatim, GRPO passes through only 27%. Rollout statistics explain why, erroneous pseudo-label words rarely appear in on-policy rollouts, so group-relative advantages stay aligned with the clean transcript, whereas SFT forces them into the policy's support and collapses the reward variance GRPO relies on.
| Comments: | submitted to ICASSP 2027 |
| Subjects: | Computation and Language (cs.CL); Audio and Speech Processing (eess.AS) |
| Cite as: | arXiv:2610.32572 [cs.CL] |
| (or arXiv:2610.32572v1 [cs.CL] for this version) | |
| https://doi.org/10.48550/arXiv.2610.32572 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Shashi Kumar [view email]
[v1]
Fri, 23 Oct 2026 13:00:46 UTC (63 KB)
来源：arXiv cs.CL · arxiv.org