Scholarlink

A voice-to-voice translation system using LSTM networks is proposed for Hindi, Bengali, Tamil, and Telugu, integrating ASR and TTS for real-time communication. Performance metrics like BLEU, WER, and MOS show significant improvements, with a BLEU score of 0.75 for Hindi, surpassing tools like Google Translate. Translation times range from 1.5 s for Hindi to 2.0 s for Tamil. User satisfaction is high for Hindi, with room for improvement in Tamil. The study highlights dialect variations’ impact on accuracy, stressing the need for diverse datasets, and opens new opportunities for multilingual voice translation.

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

Advancements in Voice Translation: Evaluating LSTM Networks for Multilingual Applications

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • B. Rajalakshmi,
  • B. Mehda,
  • Yagya Raj Bhatt,
  • Prakash Dhami,
  • Shivraj karavinakopp,
  • Devasheesh

摘要

A voice-to-voice translation system using LSTM networks is proposed for Hindi, Bengali, Tamil, and Telugu, integrating ASR and TTS for real-time communication. Performance metrics like BLEU, WER, and MOS show significant improvements, with a BLEU score of 0.75 for Hindi, surpassing tools like Google Translate. Translation times range from 1.5 s for Hindi to 2.0 s for Tamil. User satisfaction is high for Hindi, with room for improvement in Tamil. The study highlights dialect variations’ impact on accuracy, stressing the need for diverse datasets, and opens new opportunities for multilingual voice translation.