https://dl.ucsc.cmb.ac.lk/jspui/handle/123456789/4897 2026-04-06T17:59:04Z https://dl.ucsc.cmb.ac.lk/jspui/handle/123456789/4906 Title: Sum-up AI: Video Summarization Tool for Low-resource Language (Sinhala) Authors: Madarasinghe, J P; Mayunika, M R; Sooriyaarachchi, S D Abstract: Abstract Video content summarization focuses on extracting the relevant information from video data. It is very important to address the challenge of information overload, faced by users on platforms like YouTube and Vimeo. Different approaches are introduced for video content summarization using techniques such as Deep Neural Network (DNN), Natural Language Processing (NLP), Machine Learning (ML), Computer Vision, and Speech Recognition. But there exists a gap between summarizing the Sinhala language or mixed language (Sinhala + English) data. We proposed a solution to fill this gap by introducing a new platform for summarizing video content data called Sum-up AI, especially in the Sinhala language. In there, we experimented can we provide a contextually accurate summary of a video using the transcription of the video content, enabling users to grab the main idea of the entire video without wasting time to go through the whole content. For this experiment, we used three main modules: Google Transcriber API, fine-tuned sinmT5 model, and fine-tuned sinBERT model. Furthermore, our evaluation results show that the fine-tuned sinmT5 model generates summaries for a given video (only providing the input text) with 7.05 of average score out of 10, while the fine-tuned sinBERT model which is used for category classification, achieved 74% of accuracy. Findings of this research indicate that Sum-up AI provides moderately accurate summaries of Sinhala video content and serves as a good starting point for further research in the area of Sinhala video content summarization. 2025-06-30T00:00:00Z https://dl.ucsc.cmb.ac.lk/jspui/handle/123456789/4905 Title: Sinhala Speech-to-Speech Chatbot Using Deep Learning Approaches Authors: Weerakoon, T.V.R.; Nayanathara, K.K.S.; Harischandra, L.I.L. Abstract: Abstract This research presents the development of an advanced Sinhala speech-to-speech chatbot designed to bridge the gap in digital accessibility for native Sinhala speakers. Despite the rapid advancements in conversational AI systems, low-resource languages like Sinhala remain underrepresented, limiting the ability of native speakers to interact with technology in their own language. Addressing this critical gap, this study proposes an end-to-end solution that seamlessly integrates Automatic Speech Recognition (ASR), Natural Language Understanding (NLU), and Text-to-Speech (TTS) synthesis, enabling real-time, voice-based communication in Sinhala. The system leverages state-of-the-art deep learning techniques to achieve high accuracy and robustness. For ASR, transfer learning is employed to fine-tune the Wav2Vec2-BERT model on a 40-hour Sinhala speech dataset, achieving remarkable improvements with a Word Error Rate (WER) of 1.79% and a Character Error Rate (CER) of 0.33%, surpassing existing Sinhala ASR systems. The chatbot component utilizes a Retrieval-Augmented Generation (RAG) approach, combining the strengths of Large Language Models (LLMs) with dynamic knowledge retrieval to deliver context-aware and accurate responses in Sinhala. The TTS module, powered by the Variational Inference TTS (VITS) model, generates natural-sounding Sinhala speech, achieving a Mean Opinion Score (MOS) of 4.62 for intelligibility and 4.18 for naturalness in male voices, and 4.24 for intelligibility and 4.07 for naturalness in female voices. The proposed system addresses a significant gap in voice-based human-computer interaction for Sinhala speakers, with applications spanning education, accessibility, and digital services. By combining cutting-edge ASR, RAG-powered chatbot intelligence, and high-quality TTS, this research not only advances the field of NLP for low-resource languages but also sets a benchmark for future developments in multilingual speech technologies. The modular architecture and methodologies developed in this study provide a foundation for extending similar solutions to other underrepresented languages, fostering greater inclusivity in the digital age. 2025-06-30T00:00:00Z https://dl.ucsc.cmb.ac.lk/jspui/handle/123456789/4904 Title: Store Recommendation and Route Planning System for Improving Shopping Experience of Users Authors: Vandebona, B; Gunawardana, H; Satharasinghe, P Abstract: Abstract Grocery shopping often presents a frustrating experience for customers due to factors such as price discrepancies, inventory inaccuracies, and the lack of personalized assistance. Existing methods, such as store subscriptions and advertisements, are limited in their ability to deliver tailored, timely information. Furthermore, current applications require users to manually browse multiple marketplaces to locate desired products, lacking automated recommendations based on shopping lists. Addressing this gap, this undergraduate project proposes a store recommendation system designed to enhance the shopping experience. The system targets four main objectives: (1) recommending the most cost-effective stores with efficient routes for purchasing a full shopping list, (2) identifying purchase patterns and predicting future shopping needs, (3) optimizing travel paths considering dynamic conditions, and (4) proposing standardized methods for store data collection. To achieve these goals, several algorithms were implemented: greedy heuristics, branch and bound, beam search, and exact optimization methods for store selection; an adaptive genetic algorithm combined with A* search for route planning; and a Singular Value Decomposition (SVD)-based model for personalized item recommendations. Experimental results demonstrated that optimized store suggestion algorithms delivered near-optimal results within acceptable time constraints, and that route planning methods effectively reduced travel time. Challenges related to real-time inventory data acquisition are also discussed, alongside proposals for future refinements to improve data accuracy and recommendation quality. This work highlights the potential of integrated recommendation and optimization systems to significantly streamline the grocery shopping experience and sets a foundation for future real-world deployments 2025-06-30T00:00:00Z https://dl.ucsc.cmb.ac.lk/jspui/handle/123456789/4903 Title: PoHMChain: A Sustainable Cryptocurrency Platform with Proof of Human Mobility Consensus Algorithm D.M.H.P. Dissanayake M.S.W. Salgado S.N.S. Wickramasinghe 2025 Authors: Dissanayake, D.M.H.P.; Salgado, M.S.W.; Wickramasinghe, S.N.S. Abstract: Abstract Along with the positive impact, blockchain technology also has negative impacts on the environment, climate, and energy consumption. Algorithms like Proof of Work was tend to consume a great amount of energy, hence alternative methods such as Proof of Stake, Proof of Space have come into play. This dissertation focused on the newly introduced human mobility-based consensus algorithm known as Proof of Human Mobility and its practical implementation. This sustainable, proof-based, leader selection algorithm uses human mobility as the trust factor. A desktop application was implemented for blockchain connectivity, along with a mobile application, which is used to generate tickets proving human mobility. Furthermore research and evaluations were carried out in the following areas. In regard to choosing a better short-range communication method, the practical limitations of the integration of Bluetooth, Near Field Communication, and Wi-Fi Direct is discussed, and an alternative method to address those issues is given for the implementation. To make the Proof of Human Mobility (PoHM) more scalable evaluation were carried out on 8 different scenarios and calculated the average decreasing percentage and transaction per second metrics. Brotli compression with a validator group mechanism had the lowest average decreasing percentage of 26% which showcased the best strategy among all scenarios to work with, when the number of nodes increases. To identify the optimal sensor set to detect fraudulent movements evaluation were conducted on 10 sensor combinations using Random Forest Classifier with captured smartphone sensor data. The GPS + Accelerometer combination achieved 91.73% accuracy, selected as the optimal set for its higher accuracy and lower power consumption. 2025-06-28T00:00:00Z