Abstract

Converting text to its correct phonetic representation is the first stage in developing a high-quality Text to Speech synthesizer. One approach would be to discover all the rules for converting graphemes to phonemes and another would be to generate a comprehensive dictionary of words in their phonetic forms. Both approaches are necessary but none of them is feasible on its own. In order to create a comprehensive dictionary it is necessary to analyze large text corpora convert all words to their phonetic representation and then insert them into the dictionary. That is an enormous undertaking. However there are always exceptions to every rule and it would be impossible to write such laws comprehensively in Bangla and other languages. To get the most out of text-to-phonemic conversion its best to use both of these approaches together. This paper explains how the Bangla Text to Speech Synthesis System uses a hybrid technique like this. A formant-based voice synthesizer previously detailed