རྒྱབ་ཁུངས། / Background
System Architecture & Information
About the NLP Tool
The Dzongkha-English Machine Translation (MT) system was developed using Meta's NLLB model. It was trained on a parallel corpus of approximately 700,000 sentences. The Dzongkha Text-to-Speech (TTS) system was trained on a minimal corpus. All corpora were provided by the Department of Culture and Dzongkha Development (DCDD).
The models were initially fine-tuned by IT interns from CST under the guidance of DSAID, GovTech. We are continuously working to improve the models as more corpora become available. The user interface was developed and models were integrated by three interns from GCIT: Sonam Yangzom, Sujal Nepal, and Tashi Norbu Dema.
⚠️ Caution While Using the Model: This translation system is currently intended for testing purposes only. While the quality of general translation is significantly better than existing Dzongkha translation systems, it may still contain errors and inaccuracies. Therefore, we advise users to verify the output and consult human experts where critical or official accuracy is required.
About the Dzongkha LLM Chatbot
The Dzongkha Conversational AI is a platform designed to support interactions in Dzongkha through both text and voice. It allows users to communicate naturally, providing translation, dialogue, and speech-based responses. The system integrates Automatic Speech Recognition (ASR) to convert spoken input into text, translation services via WSO2 for bilingual communication, and multiple AI models including Trinity Large Preview, DeepSeek LLM, and Qwen to generate intelligent responses. Additionally, responses can be converted into speech using Text-to-Speech (TTS), enabling a full voice-based conversational experience. The platform's frontend is built with Bootstrap for responsive design, while the backend uses Flask and Python, supported by advanced AI and speech processing technologies.
The platform was developed by Kelzang Dorji (NIIT University), under the guidance of the Data Science and AI Division. The development focused on combining cutting-edge language models, translation services, and speech processing to create a system capable of handling Dzongkha conversations efficiently. The project showcases the integration of AI technologies for local language support and research purposes.
⚠️ Usage Disclaimer: This system is intended solely for testing and research purposes. Outputs may contain translation errors or incomplete responses. Users are advised to verify critical information independently and consult experts where accuracy is essential, as the system is not guaranteed to provide fully reliable or production-ready translations.