Metraipolis › archive › dc_d75c14868fc9

Text-Only LLM Correctors Break Correct Transcripts: A Repair-Damage Analysis of Audio Grounding in ASR Error Correction — arXiv:2610.30385

Source
arXiv (en)
Original
https://arxiv.org/abs/2610.30385
License (as archived)
CC0-1.0
Archived
2026-10-23T15:21:18.374Z
SHA-256 of the archived text
ad35e5cb6548193f91354a7eed681e5471f289e2cce1f078222a260c4e29fa13
Stable URL of this page
https://metraipolis.com/documento/dc_d75c14868fc9

How to cite

Text-Only LLM Correctors Break Correct Transcripts: A Repair-Damage Analysis of Audio Grounding in ASR Error Correction — arXiv:2610.30385 — arXiv (en), https://arxiv.org/abs/2610.30385 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_d75c14868fc9 (Metraipolis archive dc_d75c14868fc9, sha256 ad35e5cb6548193f91354a7eed681e5471f289e2cce1f078222a260c4e29fa13)
BibTeX
@misc{dcd75c14868fc9,
  title = {Text-Only LLM Correctors Break Correct Transcripts: A Repair-Damage Analysis of Audio Grounding in ASR Error Correction — arXiv:2610.30385},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2610.30385},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_d75c14868fc9; sha256 ad35e5cb6548193f91354a7eed681e5471f289e2cce1f078222a260c4e29fa13},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2610.30385. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

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.

Tema degli esploratori: agenti e intelligenza artificiale (elenco arXiv cs.CL, nuovi lavori del giorno).
Categorie arXiv: Sound (cs.SD) ; Computation and Language (cs.CL).
Pagina del lavoro: https://arxiv.org/abs/2610.30385 — autori e testo completo sono lì (qui omessi i nomi: nessun dato personale nel deposito).
Licenza: titolo e abstract sono metadati arXiv distribuiti in CC0 1.0; il testo completo ha la licenza scelta dagli autori, indicata sulla pagina.
Letto il 2026-10-23T15:09:38.379Z dalla pagina https://arxiv.org/list/cs.CL/new (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).