Scholarlink

This study addresses communication challenges between non-English-speaking traders and their business partners by proposing a Mooré-to-English and English-to-Mooré voice translation mobile app using Artificial Intelligence. One of the main goals of this project is to overcome the language barrier between non-English-speaking traders and their business partners. After data collection, several algorithms were tested on the initial dataset for voice translation from Mooré to English and English to Mooré. The best accuracy, 79.92%, was achieved by combining the architecture of CNN+BiLSTM on our initial dataset. By using data augmentation techniques, CNN+BiLSTM achieved an accuracy of 96.25%. Our contribution in this paper is twofold: first, we provide a dataset for the scientific community. This dataset contains recordings of basic greetings in the Mooré language. This is an important step for studies on low-resource languages like Mooré and contributes to reducing the data scarcity challenge for local languages. The second contribution concerns the Mooré speech recognition model using deep learning techniques, which can help develop further studies related to our local languages.

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

Greeting Voice Translation Using Deep Learning for Low Resource Language: Case of Mooré into English and English into Mooré

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Go Issa Traoré,
  • Borlli Michel Jonas Some,
  • Hamado Kongo,
  • Ozias Bombiri,
  • Rachid Gaetan Nabolle

摘要

This study addresses communication challenges between non-English-speaking traders and their business partners by proposing a Mooré-to-English and English-to-Mooré voice translation mobile app using Artificial Intelligence. One of the main goals of this project is to overcome the language barrier between non-English-speaking traders and their business partners. After data collection, several algorithms were tested on the initial dataset for voice translation from Mooré to English and English to Mooré. The best accuracy, 79.92%, was achieved by combining the architecture of CNN+BiLSTM on our initial dataset. By using data augmentation techniques, CNN+BiLSTM achieved an accuracy of 96.25%. Our contribution in this paper is twofold: first, we provide a dataset for the scientific community. This dataset contains recordings of basic greetings in the Mooré language. This is an important step for studies on low-resource languages like Mooré and contributes to reducing the data scarcity challenge for local languages. The second contribution concerns the Mooré speech recognition model using deep learning techniques, which can help develop further studies related to our local languages.