Speaker independent sinhala speech to text SMS application for mobile phones

Abstract

Speech recognition is one of the most discussed topics by researchers in recent years. Because of the limitless applications and the competition of making more user-friendly systems, lots of researchers put their effort on speech recognition system developments. There are lot of applications have already developed for English language. But for the languages like Sinhala, Hindi, Tamil are still at their preliminary stages. The main purpose of this study was to develop a speaker independent automatic speech recognition android application for Sinhala language. Sinhala is the native language for Sri Lankans and they are the only people who speak Sinhala Language. So this study will create a great opportunity for the Sri Lankans to build their own speech recognition applications using and enhancing this language model. The products ofthe study include frequently used SMS phrases speech corpus in Sinhala language which can be used in sending SMS. A survey has conducted among university students in order to collect frequently used SMS phrases. Those data will be used to create and implement the speech recognition system. Since the speech recognition training task is a time consuming one and the time limitation for the project restrict the size of the SMS phrases corpus to a limited one. At the moment audio recordings were only taken from one female and a one male and more recordings needed to build more accurate speech model. The System was implemented using a HMM toolkit call CMUsphinx. CMUSphinx toolkit is a leading open source speech recognition toolkit with various tools used to build speech applications. CMU Sphinx toolkit has a number of packages for different tasks and applications. Pocketsphinx is one ofthe tools that support Android operating system which comes under CMUSphinx. Pocketsphinx tool used to create a speech model that can be used in various applications. To build the speech model it needs audio recordings of text and corresponding text. Once the model created it can be used in various applications. The main aim of this project is to build an accurate speech recognition model for Sinhala language that can be used in Android operating system.