Scholarlink

An innovative framework for Automatic Speech Recognition (ASR) and language translation that combines the Wave2Vec transformer model with a post-processing stage utilizing the LLAMA 2 language model (LLM). The ASR component efficiently converts spoken audio into text, while LLAMA 2 is fine-tuned to improve grammatical accuracy, punctuation, and overall transcript clarity. By comparing the word error rates (WER) of the initial ASR output to the refined transcripts, the results demonstrate a significant reduction in errors, showcasing the effectiveness of this integrated approach.

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

Multilingual ASR Using LLAMA 2 Post-processing and Machine Translation

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • R. Geetha Rajakumari,
  • D. Karthika Renuka,
  • L. Ashok Kumar,
  • Ashfaqe Ahmed,
  • M. Mohammed Nihmathullah,
  • S. Nithin Kumar,
  • V. Rohith,
  • S. Yuvarraj

摘要

An innovative framework for Automatic Speech Recognition (ASR) and language translation that combines the Wave2Vec transformer model with a post-processing stage utilizing the LLAMA 2 language model (LLM). The ASR component efficiently converts spoken audio into text, while LLAMA 2 is fine-tuned to improve grammatical accuracy, punctuation, and overall transcript clarity. By comparing the word error rates (WER) of the initial ASR output to the refined transcripts, the results demonstrate a significant reduction in errors, showcasing the effectiveness of this integrated approach.