Multifarious Tamil Text Processing for a Speech Synthesizer

Abstract

A Text-To-Speech (TTS) Synthesizer provides auditory information access by newlineprocessing the raw natural language text into equivalent sound unit transcriptions.TTS offers several benefits to the people with visual and reading impairments,people who have literacy difficulties, mobile application users, online learners, and many more. Apparently, a TTS in one s own native language increases engagement of people with the digital world. The proposed thesis concerts about the development of a Text Processing module for a TTS in Tamil language, which is one of the longest-surviving classical languages with 76 million speakers around the world. In this digital era, abundant of digital text is available even in regional newlinelanguages such as Tamil in terms of blogs, wiki pages, user and movie reviews etc. which we call as multifarious text. Most of the existing Tamil Natural Language Processing (NLP) algorithms in literature recede from processing this multifarious text. Moreover, idiosyncrasies of letter to sound unit mapping in Tamil language escalates the complexity for pronunciation transcription generation. The prime conception of the proposed thesis is to resolve these challenges and hence we have newlinedeveloped (i) a text normalizer which converts words with non-orthographic symbols into orthographic words using a semiotic decision list with an accuracy of 97.89% on the test set; (ii) an archetype of a sentiment analyser to aid prosody detection using an aggregative linguistic score newline