arXiv：cs.AI· Shiran Liu, Gaohui Xu, Sergio Burdisso, Thibault Ba\~neras-Roux, Shashi Kumar, Petr Motlicek·· 6 小时前
基于 N-Best 假设的 oracle 引导微调：面向生成式 ASR 后编辑的选择或重写
Select or Rewrite: Oracle-Guided Fine-Tuning for Generative ASR Post-Editing from N-Best Hypotheses
AI 导读
研究者提出 oracle 引导微调方法，将 ASR 的 N-best 列表组织为纠错、正则化和难负例三类数据池，让大语言模型学习"选择或重写"的混合策略，而非仅靠提示词纠正。该方法在约 3B 参数规模上于专有数据集取得最明显增益，但在 CommonVoice 上结论不明确，作者推测公开朗读文本可能与 LLM 预训练数据重叠，掩盖了真实的后编辑增益。
正文
Abstract:Large language models can post-edit ASR by using linguistic context to compare N-best hypotheses and rewrite the transcript when needed, often improving WER without retraining the recognizer. However, prompt-only correction is brittle: it may over-edit when top-1 is correct and hallucinate when the reference is absent, and often needs very large models. We propose oracle-guided fine-tuning that turns N-best lists into supervision via three curated pools: corrections (reference in the list but not top-1) to teach selection, regularizers (top-1 correct) to discourage unnecessary rewrites, and hard negatives (reference absent) to suppress hallucination. Fine-tuning on this data provides more effective supervision than random subset fine-tuning for learning a hybrid select-or-rewrite policy, with the clearest gains from around 3B parameters on the proprietary dataset but less conclusive results on CommonVoice, a divergence suggesting that public read-speech text may overlap with LLM pretraining and mask genuine post-editing gains.
| Comments: | 7 pages, 4 figures, 2 tabl |
| Subjects: | Computation and Language (cs.CL); Artificial Intelligence (cs.AI) |
| Cite as: | arXiv:2610.15056 [cs.CL] |
| (or arXiv:2610.15056v1 [cs.CL] for this version) | |
| https://doi.org/10.48550/arXiv.2610.15056 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Shiran Liu [view email]
[v1]
Sat, 10 Oct 2026 16:48:27 UTC (383 KB)
来源：arXiv：cs.AI · arxiv.org