跳到正文
arXiv cs.CL· Abhishek Kumar, Aashraya Sachdeva·· 2 天前AI 评分31

RECOVER:用多假设智能体编排实现 ASR 实体纠错

RECOVER: Robust Entity Correction via agentic Orchestration of hypothesis Variants for Evidence-based Recovery

AI 导读

RECOVER 是一个面向 ASR 后处理的智能体实体纠错框架,利用多个识别假设作为证据检索实体,再由 LLM 在约束下提出修正并经确定性护栏与 LLM 评审过滤,必要时触发反馈轮次找回首次遗漏的实体。

正文 · 原文

View PDF HTML (experimental)

Abstract:Entity recognition in Automatic Speech Recognition (ASR) is challenging for rare and domain-specific terms. In domains such as finance, medicine, and air-traffic control, these errors are costly. If an entity is entirely absent from the ASR output, text-only post-correction has little signal to recover it. To address this, we introduce RECOVER, an agentic post-ASR entity-correction framework. It leverages multiple hypotheses as evidence from ASR, retrieves relevant entities, proposes Large Language Model (LLM) correction under constraints, filters them through deterministic guardrails and uses an LLM critic that judges each surviving edit and can trigger a targeted feedback round to recover an entity the first pass missed. The hypotheses are used by one of several strategies, namely 1-Best, Entity-Aware Select, Recognizer Output Voting Error Reduction (ROVER) Ensemble, LLM-Select and Adaptive Routing. Evaluated across five diverse datasets over a Whisper-large-v3 baseline, it achieves up to 49% relative reduction in entity-phrase word error rate (E-WER) and increases entity recall by up to 14 percentage points while preserving the overall word error rate (WER), with LLM-Select the strongest strategy.
Comments: Accepted to EMNLP 2026 Industry Track
Subjects: Computation and Language (cs.CL); Audio and Speech Processing (eess.AS)
Cite as: arXiv:2603.16411 [cs.CL]
  (or arXiv:2603.16411v2 [cs.CL] for this version)
  https://doi.org/10.48550/arXiv.2603.16411

arXiv-issued DOI via DataCite

Journal reference: Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing: Industry Track, pages 2939-2951, Budapest, Hungary
Related DOI: https://doi.org/10.18653/2026.emnlp-industry.173

DOI(s) linking to related resources

Submission history

From: Abhishek Kumar [view email]
[v1] Tue, 17 Mar 2026 11:48:12 UTC (96 KB)
[v2] Fri, 13 Nov 2026 15:38:12 UTC (42 KB)

来源:arXiv cs.CL · arxiv.org