纯文本 LLM 纠错器会破坏正确转写：ASR 纠错中音频 grounding 的修复-损伤分析

Text-Only LLM Correctors Break Correct Transcripts: A Repair-Damage Analysis of Audio Grounding in ASR Error Correction

AI 导读

研究将 ASR 纠错带来的 WER 变化精确拆分为修复 token、损伤 token 和插入，发现 Gemini 2.5 Flash 仅用文本纠正自身草稿时，会破坏 13-32% 原本正确的转写，加入音频后降至 1-3%。音频使损伤在两个设置下均下降，但修复在自身草稿上减少、在独立识别器输出上增加。医疗听写场景中保留的插入抵消了净 WER 收益，而很少改动的开源 Voxtral 未出现该模式。

正文 · 原文

View PDF HTML (experimental)

Abstract:LLM correctors of ASR output are judged by aggregate WER, which hides what they fix and what they break. We decompose every WER change exactly into repaired tokens, damaged tokens and insertions, and use this to ask what the source audio adds to a corrector. We compare correction with and without the audio, for the model's own draft and for the output of an independent recogniser, on three corpora. With Gemini 2.5 Flash, text-only correction of its own draft breaks 13-32% of already-correct transcripts; with the audio, 1-3%. Damage falls in both settings. Repair falls against the own draft and rises against the recogniser. On medical dictation, kept insertions cancel the gain in net WER. The open-weight Voxtral, which rarely edits, does not show this pattern. Prompts differ between the two arms, so the contrasts bound the effect of the audio and do not isolate it.

Comments:	Submitted to IEEE ICASSP 2027
Subjects:	Sound (cs.SD); Computation and Language (cs.CL)
Cite as:	arXiv:2610.30385 [cs.SD]
	(or arXiv:2610.30385v1 [cs.SD] for this version)
	https://doi.org/10.48550/arXiv.2610.30385 arXiv-issued DOI via DataCite (pending registration)

Submission history

From: Kirill Borodin [view email]
[v1] Wed, 21 Oct 2026 19:45:31 UTC (22 KB)

来源：arXiv cs.CL · arxiv.org