Scholarlink

This paper explores the integration of automatic speech recognition (ASR) with large language models (LLMs), aiming to validate the effectiveness of this combination, particularly for automatic post-editing (PE) tasks. Initially, we investigate the use of LLMs for ASR PE error correction, performing second-pass rescoring on the output transcriptions generated by the ASR system, using both N-best decoding hypotheses and lattices. Subsequently, we examine the combination of ASR outputs from various systems using LLMs, addressing a classic system combination task. Experimental results demonstrate that LLMs can offer substantial assistance in automatic PE.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Automatic Post-editing of Speech Recognition System Output Using Large Language Models

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Sheng Li,
  • Jiyi Li,
  • Yang Cao

摘要

This paper explores the integration of automatic speech recognition (ASR) with large language models (LLMs), aiming to validate the effectiveness of this combination, particularly for automatic post-editing (PE) tasks. Initially, we investigate the use of LLMs for ASR PE error correction, performing second-pass rescoring on the output transcriptions generated by the ASR system, using both N-best decoding hypotheses and lattices. Subsequently, we examine the combination of ASR outputs from various systems using LLMs, addressing a classic system combination task. Experimental results demonstrate that LLMs can offer substantial assistance in automatic PE.