Disha: A Bilingual Humanoid Virtual Assistant

• Published in: 2024 6th International Conference on Electrical Engineering and Information & Communication Technology (ICEEICT) pp. 457 - 462
• Main Authors: Ullah, Md. Rifat, Mahbub, Md. Nafees, Hakim, Md. Azizul, Sultana, Yeasmin, Alam, Iftakhar
• Format: Conference Proceeding
• Language: English
• Published: IEEE 02.05.2024
• Subjects: Automatic Speech Recognition; Banking; Chatbots; Communications technology; Human-Computer Interaction; Humanoid robots; Language Model; Natural Language Processing; Natural Language Understanding; RASA; Real-time systems; Speech processing; Text-to-Speech; Virtual Assistant; Virtual assistants
• ISSN: 2769-5700
• DOI: 10.1109/ICEEICT62016.2024.10534321

Researchers and tech giants have developed AI-based virtual assistants that can answer basic questions, largely in English. As People are consumed by smart devices in modern life, a virtual assistant can be a game-changer. Surprisingly, no work has been done to adapt AI-based virtual assistants for the Bengali language. In this paper, we introduce a humanoid virtual assistant named "Disha" designed to cater primarily to the banking sector in Bangladesh. Disha is capable of answering customers' financial service-related questions and conducting real-time transactions. Initially, we developed a web-chat system using RASA and trained models for automatic speech recognition (ASR) and text-to-speech (TTS) that operate in both Bengali and English separately. This project not only serves the current banking needs but also contributes to future research, potentially reducing the reliance on human agents and replacing traditional IVR and chatbots in the banking sector.