DEVELOPING A MOBILE APPLICATION FOR MACHINE TRANSLATION USING WAV2VEC 2.0 AND MBART

Abstract

This thesis presents the development of a mobile application for Vietnamese-toEnglish translation by integrating advanced deep learning models: Wav2Vec 2.0 for converting speech to text and mBART for machine translation. The project addresses the demand for reliable Vietnamese-English translation tools on mobile devices, making use of both speech recognition and multilingual translation capabilities. In this system, Vietnamese speech is first transcribed into text using Wav2Vec 2.0, then translated into English with mBART. The approach achieved a BLEU score of 23.28, demonstrating competitive performance relative to the baseline Vit5 model. This result highlights the potential of the application for practical use, with future work aimed at further improving translation accuracy and expanding the dataset to support broader usage.