Scholarlink

This research paper scours into the fine-tuning of OpenAI’s Whisper and its distilled counterpart, DistilWhisper, across many models and configurations. Whisper is diagnosed for its robust automatic speech popularity (ASR) skills, at the same time as DistilWhisper offers a greater computationally fast and efficient alternative. We found the architecture and libraries utilized in these fashions, the detailed procedure of quality-tuning on good-sized datasets, and the comparative effects well-on phrases of Word Error Rate (WER) and Character Error Rate (CER). The observation consists of hyperlinks to datasets and a complete evaluation of a few of Whisper and DistilWhisper fashions to offer a vast perspective on their performance and alertness capacity.

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

Fine-Tuning OpenAI Whisper and DistilWhisper: An In-Depth Analysis

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Aman Yadav,
  • Anita Shrotriya,
  • Amit Kumar Bairwa

摘要

This research paper scours into the fine-tuning of OpenAI’s Whisper and its distilled counterpart, DistilWhisper, across many models and configurations. Whisper is diagnosed for its robust automatic speech popularity (ASR) skills, at the same time as DistilWhisper offers a greater computationally fast and efficient alternative. We found the architecture and libraries utilized in these fashions, the detailed procedure of quality-tuning on good-sized datasets, and the comparative effects well-on phrases of Word Error Rate (WER) and Character Error Rate (CER). The observation consists of hyperlinks to datasets and a complete evaluation of a few of Whisper and DistilWhisper fashions to offer a vast perspective on their performance and alertness capacity.