Pustaka Ilmiah Universitas Padjadjaran

Abstrak

Automatic Indonesian Word Text To Sound Conversion Using Finite State Transducer Of Data Flow Diagram In Speech Processing

By combining machine learning techniques of finite state transducer (FST) in four-steps of data flow diagram level-I, we can develop conversion system of word text to sound automatically speech processing that recognizing and generating the pronunciation of words as correct as the pronunced words or inversely through a transformation-based learning. Our interest here is to design in device that is able to generate Indonesian word text to sound directly. We apply transducer composition to represent the combination of linguistic, phonetic and acoustic information at various levels as found. Generally, word text decomposition proceeds for each syillable recognizing must follow the rule of Indonesian word text decomposition by spelling to word syllables that consists of consonant and nowel sequences include in diphthong syllable identification. The syllables of phonetics are converted to acoustic composition same as the finite state transducer rules. A finite-state method based on leftmost longest-match replacement is presented here for segmenting words into syllables and for converting words into phonetics, which provides possibilities for developing transducer based on their conversion rules. Futher, these word compositions are changed to sounds that are ready spoken. It is text to sound generated. It proposes is to know how the Indonesian syllables can be pronounced authentically as sounds orally. These sounds are recorded wav-formats as audio-digital files. However, to store these way-files we need an enormous hard-disk space, 3834 files with size of 344 MB.