DIFFICULTIES AND SOLUTIONS OF TURKISH TEXT TO SPEECH SYNTHESIS

Abstract

Even though numerous research studies have so far been conducted on “text-to-speech (TTS)” systems for different languages, the number of studies regarding agglutinative languages such as Turkish is relatively few. A feature of agglutinative languages is that many words and phrases can be built up from one etymon, various ablauts can be observed and therefore some problems may exist in Turkish TTS synthesis process. This study explains the development of a TTS system for the Turkish language, which uses a “Vowel-to-Vowel” phoneme method. According to this method, the system takes written text as input, makes the normalization of the text and produces Turkish sounds equivalent to given text. The implementation does not only vocalize the text, but examines it semantically and identifies its Turkish pronunciation. As a result, the difficulties with the application of Turkish TTS synthesis system were determined. Various solution methods for these problems dealing with punctuation marks, abbreviations, numeric expressions and sound alteration were generated, and hence a special Turkish TTS synthesis system which is close to human voice was developed.

Keywords

• Turkish
• Text-to-Speech Synthesis (TTS)
• Speech Synthesis
• Vowel-to-Vowel