A GRAPHEME-TO-PHONEME TRANSLATOR FOR TTS SYNTHESIS IN GREEK

摘要

n this paper is presented the algorithm of an automatic grapheme-to-phoneme translator for the Greek language. The proposed algorithm is designed to collaborate with a high quality Text-to-Speech synthesis system in Greek. The algorithm assimilates the full reading process of written text as realized by a Greek speaking person. A detailed study of the Greek language's operation, led us to the implementation of an automatic integrated system which describes its phonetic behaviour in an exact and natural way. The software that implements the algorithm has the capability to receive written text from any input (keyboard, file, screen reader, OCR system e.t.c.) and transform it to phonetic form. Afterwards the output of the algorithm is directed to the input of a concatenation-based speech synthesizer and the right pronunciation of any written text is achieved in real-time. During the reading process the software locates and distinguishes Greek written text from any foreign language words, specially written symbols, abbreviations e. t. c... and afterwards manages them in order that the flow of the reading process permits the right perception of the produced spoken messages. The most important qualification of the algorithm is the possibility to incorporate it in other Text-to-Speech synthesis systems of different technology. Finally experimental measurements indicate the successful operation of the algorithm.