aiqbench analysis
This paper is an empirical study of LLM-based correctors for speech recognition output. It splits every WER change into repaired, damaged, and inserted tokens, showing that aggregate WER hides how often correction breaks transcripts that were already right. Its main finding is that text-only correction broke 13-32% of correct transcripts versus 1-3% with audio, though the prompts differ between arms, so the audio's effect is bounded rather than isolated.
Text-Only LLM Correctors Break Correct Transcripts: A Repair-Damage Analysis of Audio Grounding in ASR Error Correction
arXiv:2610.30385v1 Announce Type: cross 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.
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.
Sound (cs.SD); Computation and Language (cs.CL)
arXiv:2610.30385 [cs.SD]
(or arXiv:2610.30385v1 [cs.SD] for this version)
https://doi.org/10.48550/arXiv.2610.30385 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Full text under the publisher's open licence · 243 words