Researchers have developed RECOVER, a new framework designed to improve entity recognition in Automatic Speech Recognition (ASR) systems, particularly for rare or domain-specific terms. This agentic system utilizes multiple ASR hypotheses as evidence, retrieves relevant entities, and employs Large Language Models (LLMs) for constrained correction. It further refines these corrections through deterministic guardrails and an LLM critic, with the ability to trigger feedback loops for missed entities. Evaluated against a Whisper-large-v3 baseline across five datasets, RECOVER demonstrated significant reductions in entity-phrase word error rate and increased entity recall without compromising overall word error rate, with the LLM-Select strategy proving most effective. AI
IMPACT Enhances the accuracy of speech recognition systems for specialized terminology, potentially improving usability in critical domains like finance and medicine.
RANK_REASON The cluster contains a research paper detailing a new framework and methodology. [lever_c_demoted from research: ic=1 ai=1.0]
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →