http://dlib.iit.ac.lk/xmlui/handle/123456789/1349 2026-04-21T07:20:44Z 2026-04-21T07:20:44Z Dias, Danoja http://dlib.iit.ac.lk/xmlui/handle/123456789/1500 2023-01-23T04:08:41Z 2022-01-01T00:00:00Z

AN Expressive Text To Speech System for Sinhala Language Dias, Danoja This research attempts to investigate creating Text-To-Speech (TTS) System for Low Resource Sinhala Language using FastSpeech2. This project uses already available TTS datasets for the Sinhala Language. This research updates the existing open-source pronunciation dictionary by adding approximately 10000 new words with their pronunciations. New Montreal Forced Aligner (MFA) models, which were not available for the Sinhala Language, were created to do the forced alignments for this research. The evaluations were done with 20 native speakers to evaluate and calculate the MOS values. The experiments with FastSpeech2 have positive results with a MOS value of 3.35 ± 0.08

2022-01-01T00:00:00Z Raterala Mudiyanselage, Sanduni http://dlib.iit.ac.lk/xmlui/handle/123456789/1499 2023-01-23T04:01:07Z 2022-01-01T00:00:00Z

VAS Service Prediction for Telco Customers Raterala Mudiyanselage, Sanduni "A mobile telecommunications company's value-added services provide clients a range of services. Annual revenue from value-added services is significant for telecommunications businesses. Value Added Services are offered by a value-added service provider externally or internally by the mobile network operator. The network operators make a significant amount of money from these services. It is vital to identify customers from the current customer base who are qualified for each VAS and recommend these services to them in order to reap the greatest benefits. The proper clients will be more likely to use the service and boost revenue for the business if it is recommended to them. These VAS will increase both the consumer base and customer satisfaction. The contented customers of a business are its most important asset, to sum up. On the other hand, a customer will be dissatisfied with the operator if he is suggested a service that he does not desire. Because the consumer will find the messages that make those suggestions burdensome and will become irate with the network operator. Therefore, it is critical to identify the right customers for each VAS. This research tries to research, design and develop accurate VAS prediction model for telecommunication customers using machine learning. The research used a dataset from a prominent Mobile Service Provider in Sri Lanka and extracted 9 important features to build the model. Different machine learning algorithms, including classical algorithms and ensemble methods, are used to build many models. Stacking, Random Forest, XG Boosting, Bagged CART were used alongside with Logistic Regression, K-nearest neighbor (KNN) and Naïve bayes (NB) algorithms for the predictive modeling and analysis. Predictive model which was built using bagging Classification and Regression Trees algorithm (CART) showed the best results among other models with accuracy of 82.97%."

2022-01-01T00:00:00Z Jalangan, Miluckshan http://dlib.iit.ac.lk/xmlui/handle/123456789/1383 2023-01-12T05:49:12Z 2022-01-01T00:00:00Z

Enhance Learning American Sign Language by Gamification: Bridging Computer Vision and Games Jalangan, Miluckshan "Sign language is the means of communication for deaf people due to their hearing impairments. Deaf people often find it difficult to adapt in society and carry out daily activities such as shopping due to communication issues. Sign language is not a popular mode of communication in the society and therefore, the deaf people will have to spend much energy or use sign language translators while conversing with non-sign language speakers. In this research, a responsive web based application is proposed with gamification approaches inbuilt to enhance the learning experience as such approaches are proven to help learn languages better in literature. Initially, a prototype spelling game was tested out with 10 participants and received mostly positive feedback. The application was tested out to detect American Sign Language (ASL) hand signs alphabets from A to Z as deaf people would often use letter spelling when referring to words that are not in the official sign languages such as current events or technical topics. The application leverages the use of computer vision and uses image classification approach to detect hand signs. A Convolutional Neural Networks (CNN) model was built and the built model scored 99.79% accuracy in identifying 29 classes - 26 Alphabets, Space, Delete and Nothing."

2022-01-01T00:00:00Z Norbertsan, Anet http://dlib.iit.ac.lk/xmlui/handle/123456789/1382 2023-01-12T05:45:04Z 2022-01-01T00:00:00Z

Code-Mix Chatbot for Customer Service Norbertsan, Anet "Chatbot is a computer program simulate human conversation. Nowadays artificial intelligence features are added to mimic the human characteristics in conversation. Specially in customer care services chatbots are used to handle queries. Natural Language Processing (NLP) techniques are the base behind the artificial conversation. After the creation of transformer, NLP field significantly evolved. Usage of more than one language in one sentence which is called Code-Mix is very common in social medias and human chats. This research is based on the code-mix chatbot which can handle customer queries in bank. Traditional machine learning and the transformer techniques are used to handle the code-mix language. In this research two translator with language identifier model and the code-mix data trained transformer model were developed and their performance was compared. The testing results has proved that the translation model with language identifier achieved more accuracy than the code-mix data trained transformer model. Code-mix dataset, code-mix language identifier and the chatbot which can handle code-mix queries related to bank customer service are the novel outputs from this research. "

2022-01-01T00:00:00Z