R?fat A?l?yan

Turkish Speech Recognition: a New Approach to Speech Recognition for Syllabified Languages R?fat A?l?yan

Nowadays, speech recognition studies are quite popular and increasing fast. New computer technologies have promoted the studies in this field. Most systems are generally based on phoneme and word. Turkish is one of the most widely spoken, but the less studied languages. Because Turkish is an agglutinative language, the word based systems are not adequate for Turkish speech recognition. In this book, how to be designed syllable based Turkish speech recognition systems based on LTA, DTW, ANN, HMM and SVM is mentioned in detail. The summary of the systems is explained in the following. The isolated word recognition systems consist of five parts: Preprocessing, feature extraction, training, recognition and postprocessing. Preprocessing includes signal smoothing, windowing and syllable end-point detection. MFCC, LPC, parcor, cepstrum and rasta have been used in feature extraction. In training, the syllable models are generated. In recognition, the utterances are compared with the syllable models. Then, the recognized syllables are concatenated with each other. In postprocessing, the system decides whether the recognized word is Turkish or not.