Enhancing Sinhala Text-to-Speech System Using Deep Learning Techniques.

Abstract

ABSTRACT Knowledge is an important asset to all human beings to lead a successful life. Normally people gain knowledge through several ways and resources. Visually impaired people face a lot of problems throughout their lifetime since normally they gain knowledge through word of mouth, audio books, and using braille systems. They need support or assistance to carry out even their day-to-day tasks. So, it is very important to initiate necessary steps to help them using the prevailing technologies. So that they can also lead a good life. Nearly a million in Sri Lanka suffer from blindness or from conditions that could lead to blindness. Blind people are unable to perform visual tasks. Most published printed works do not include braille or audio versions. There are some systems that use the OCR framework for recognition of its text, which is then synthesized through a process of TTS for languages such as English, Tamil, etc. This study focused on enhancing Text-to-Speech (TTS) technology for the Sinhala language, aiming to improve accessibility for visually impaired individuals in Sri Lanka. It tackled the challenge of adapting TTS for a low-resource language by utilizing the VAENAR model, a strategy previously successful with English, in pursuit of creating a Sinhala TTS system capable of delivering natural and intelligible speech. Despite confronting substantial obstacles, including significant computational requirements and the inadequacy of the model to produce clear speech in both English and Sinhala, the research provided important directions for future TTS development. The outcomes underscored the critical need for tailored deep learning approaches, enhanced linguistic data collection, and stronger collaborative networks within the academic and research communities. These elements are vital for crafting TTS technologies that are accessible and useful to visually impaired users and broadly beneficial across various linguistic groups. This work lays a foundation for future innovations in TTS systems, advocating for more inclusive and effective solutions for the Sinhala language and other low-resource languages, thereby offering significant contributions to the field and its potential impact on society