BILI: A Domain-Specific Reception Assistant Robot for Bilingual and Regional Language Interaction in Bangladesh

Abstract

This project addresses digital inequality and linguistic exclusion in Bangladeshi academic environments by developing a Bilingual University Assistant Robot (BILI) and a supporting speech dataset named BRADS. Although Bangla is one of the most widely spoken languages globally, current digital assistants primarily support only standard Bangla, marginalizing millions of regional dialect speakers in educational institutions. To bridge this gap, we introduce BRADS—a curated audio dataset containing 2,439 recordings of 298 frequently used university-related words, including 233 regional and 65 standard terms, collected from 85 native speakers across all eight divisions of Bangladesh. BILI leverages this dataset to enable real-time dialect recognition using a CNN–BiLSTM hybrid model. The robot features natural language processing, MFCCbased speech feature extraction, and a dialect-tuned text-to-speech (TTS) module, allowing it to interact fluently in both Bangla (standard and regional dialects) and English. Optimized for low-latency edge deployment on the NVIDIA Jetson Xavier NX, BILI offers responsive, multilingual support in real time. It assists students and visitors with educational queries, navigation across campus buildings using Dijkstra’s algorithm, and interactive voice-visual responses through a smart interface. Field deployment at Independent University, Bangladesh (IUB) showed over 1,900 successful interactions in Bangla, English, and regional dialects, achieving high precision and user engagement. Future developments include expanding BRADS to support conversational and emotional speech and integrating region-specific voice synthesis for even more personalized experiences. BILI demonstrates a scalable solution for inclusive, voice-enabled university services in low-resource and linguistically diverse academic settings.