Formant synthesis based TTS application for Sinhala Unicode

Abstract

Text to speech or speech synthesis is the way of producing the human voice artificially. It is one of the main research areas for last five decades. Consequently researchers came across with different set of techniques to implement speech synthesis applications. Formant synthesis is one of such a speech synthesis technique and it does not use human speech samples at runtime; instead synthesized speech is created by using an acoustic model. Since this technique does not require pre recorded voices, it requires minimum disk space and runtime resources. Furthermore formant synthesis based text to speech applications can be configured easily for different voices like male and female. Other speech synthesis techniques do not have this flexibility. Furthermore they are heavy-weighted too and cannot be used in smaller devices like smart phones. There are few formant synthesis based speech synthesizers available to use but all of them are for English and widely used languages. eSpeak is one of such a free open source formant synthesis based speech synthesizer framework written in C++ with multilingual support. In-built Sinhala spoken support is not there but eSpeak framework provides an API for adding new languages. In this paper I present a Sinhala TTS tool called SieSpeak and it is based on this eSpeak API. SieSpeak is capable of reading not only the Sinhala texts, but also abbreviations, numbers, date\time and much more. eSpeak framework is delivered with English accent and its acoustic model produces a robotic voice. These two problems also addressed in this tool as part of voice naturalization. I would like to dedicate this tool to the visually impaired Sinhala community in Sri Lanka. They are demanding these types of tools, because they do not have a proper way of reading electronic articles. The tool presented here is ideal for them to read Sinhala articles if they are capable of using computers. Also this is a really valuable tool for the people who are hard to find the time for reading. They can use this tool for reading Sinhala articles while doing something else. Keywords: Sinhala TTS, Sinhala Unicode, formant synthesis, eSpeak