📄 中文摘要
自动语音识别（ASR）系统在处理特定领域短语（如命名实体）时常常出现误识别，可能导致下游任务的严重失败。近期，基于大型语言模型（LLMs）的命名实体纠正方法逐渐兴起，但这些方法尚未充分利用LLMs固有的复杂推理能力。为了解决这一问题，提出了一种新颖的检索增强生成框架，用于纠正ASR中的命名实体错误。该框架包括两个关键组件：首先是用于命名实体识别的重述语言模型（RLM），然后通过音素级编辑距离进行候选项检索；其次是一个新颖的自学推理模型，能够提升命名实体纠正的准确性和效率。
基于自适应思维链的检索增强自学推理模型用于自动语音识别命名实体纠正
📄 English Summary
Retrieval-Augmented Self-Taught Reasoning Model with Adaptive Chain-of-Thought for ASR Named Entity Correction
End-to-end automatic speech recognition (ASR) systems often misrecognize domain-specific phrases such as named entities, leading to significant failures in downstream tasks. Recently, a new class of named entity correction methods based on large language models (LLMs) has emerged, yet these approaches have not fully leveraged the sophisticated reasoning capabilities of LLMs. To address this gap, a novel retrieval-augmented generation framework for correcting named entity errors in ASR is proposed. This framework consists of two key components: (1) a rephrasing language model (RLM) for named entity recognition, followed by candidate retrieval using phonetic-level edit distance; and (2) a novel self-taught reasoning model that enhances the accuracy and efficiency of named entity correction.