Turkish text to speech using children's voices syllables

Text to speech (TTS) shortly means to convert a written text into audio signals electronically. This written text may be a text document, electronic book, or a web page. An ideal TTS system is expected to be able to process every readable text in the quality of natural human voice. In our country, text to speech studies mostly focus on the production of adult male and female voices. In this thesis, an audio database consisting of children's voices was designed so the synthesized sound is aimed to be children's voices. In voice synthesis studies, it is seen that the closest sound to naturalness was provided by concatenative voice synthesis methods. Within the scope of this thesis, a TTS system that is based on additive synthesis technique which uses binary syllable as the length of voice unit is implemented. In general, conversion of text to audio signal process consists of two main parts. In the first part, the text to be synthesized is normalized according to language rules and is divided into syllables. A hyphenation algorithm is developed for the designed system and the entered text was separated into syllables. In the second part, audio syllable signals are processed and merged so that the speech synthesizing process is performed. Although there are different techniques in processing the audio signals, they are extended and shortened based on the Synchronous Overlap and Add (SOAP) method in this thesis. The system generates syllables from the text information it receives as an input. It makes triple syllables to be produced from double syllables. Then, by using the audio files belonging to these syllables, syllables are taken from the recorded files and began to be merged. At this stage, rules determined according to the types of sounds are applied at the junction points of syllables and naturalness is tried to be created similar to the waveforms in real sound files. This naturalness has been tried to be provided by extending and shortening the beginning or end of syllables where necessary. Although the system uses simple techniques, the selected additive method is very suitable for the structure of Turkish and so produces efficient results.