Text to Speech Conversion for Odia Language

Abstract

Text-to-Speech (TTS) system is designed to generate natural and intelligible sounding speech from any arbitrary input Odia digital text. The arbitrary text can be generated from a corresponding image file that undergoes processing through an Optical Character Recognition (OCR) scheme. This project is a research work on various techniques viable to develop the segmentation phase of the OCR specifically and a corresponding TTS for Odia language. The pre-processing and segmentation phases of the OCR are thoroughly explored in our work. A scheme is contrived for extracting the atomic Odia characters from a given string of text consisting of both characters and matras. The concept of histograms has come handy in our research. An altogether new L-corner scheme is formulated to handle the exceptional cases. Extensive simulations are carried out to validate the efficacy of the proposed algorithm. Results show that the proposed scheme attains a high level of accuracy. The methodology used in TTS is to exploit acoustic representations of speech for synthesis, together with linguistic analyses of text to extract correct pronunciations and prosody in context of Odia language. Phonetic analysis to convert grapheme to phoneme is achieved using an array numbering system for the audio files comprising of pre-recorded natural human speech. Concatenation of phones and diphones generates the speech. The phoneme database creation studying the prosody of Odia language is our primary focus in this project apart from accurate and intelligible speech generation.