Automatic speech recognition and sign language translation for Turkish

In this thesis, we are working on a system that can help people with hearing impairments to communicate actively with people who are not hearing impaired. The system generally works in two steps. First, the speech is translated into text. The sign language animation or video to which the text corresponds is then shown. During the translation into text phase, Turkish Language has been analyzed and it is a phoneme-based language. With the development of industry, security, communication and robotic systems, the use of Automatic Speech recognition (ASR) based applications is increasing day by day. Along with technological developments, while ASR applications are becoming common in many languages, these applications are not much in agglutinative language groups such as Turkish, Finnish and Hungarian. Turkish has an additive morphology as a syllable structure. This structure causes a great increase in vocabulary. A phoneme and subword based recognition system has been designed for the future so that the system can detect the words other than the database approximately. In addition to classical methods, intelligent and learning methods are frequently used in this field. In this study, current Deep Learning comprehensive applications have been developed for solving the ASR problem in Turkish language. In the Speech Recognition step of our study, Deep Belief Networks (DBN), Long-Short-Term Memory Networks (LSTM) and Gated Repetitive Units (GRU) Deep Learning techniques were applied and their performances were compared. It has been seen that the most successful method is with the GRU method, where language modeling is also done with deep learning. The performance of the methods was compared against standard criteria. The study, while investigating the subject in detail about the ASR applications, also gave detailed information about the application method of the method. After the improvement made for Turkish in the step of speech definition, the word we obtained by converting the speech into writing was found to correspond to the sign in Sign Language and translated into Turkish Sign Language with these sign videos. This study, which does not appear in the literature for Turkish, converts Turkish Speech to Turkish Sign Language, is thought to be a pioneering work to increase performance in Turkish speech recognition systems and facilitate the lives of hearing impaired people.