An Intelligent Software Tool for Audio Signal-based Hinglish to English Language Translation

Abstract

Most of the current Indian generation speaks/writes using Hinglish on various social media platforms. It is difficult to understand and interpret Hinglish using existing computational linguistic tools mainly based on English. Hence, there is a need to provide automated Hinglish to English Language Translation tools to integrate with existing NLP-based applications like hate speech detection. Currently, such complex language translations are manually intensive and require domain experts. Also, there is a scarcity of standard datasets in this domain for carrying out helpful research. In this research, we propose an intelligent AI-based translator for translating audio data signals from Hinglish to English, which can be integrated into various NLP-based applications in the Indian Context. We also created a large-scale customized annotated Hinglish to English dataset containing 27135 entries of size 4.121MB. After extensive experimentation, we received the highest BLEU (Bi-Lingual Evaluation Understudy) score of 92.20 on our customized Hinglish to English dataset, while on the original PHINC dataset, we found a BLEU score of 70 approx. We also developed a Python TKinter-based user-friendly GUI application that takes audio signal/voice as input and provides English-translated text as output. � 2023 IEEE.