HUMAN AND ARTIFICIAL RECOGNITION OF SPEECH DISORDERS OF CHILDREN
One of the most important and provocative characteristics of a Computer Based Speech Therapy System (CBST) is to provide real time feedback – an ability traditionally reserved for humans. That is why, artificial recognition of speech disorders (and of severity of speech disorders) has become an important and dynamic research field. Consequently, the development of a pronunciation quality evaluation framework and the integration of this module in Logomon – the first CBST for Romanian language – has become a priority for our team. In this paper we present both theoretical and practical related issues such as: acquisition of data, human scoring, and human scoring consistency. A brief literature review is presented, exploring the recent work in the area. The consistency of human evaluation of pronunciation is evaluated on both inter- and intra-rater level using classical and extended (for more than two raters) correlation coefficient. The obtained results can be seen as an intermediary step in order to archive a real time, quality automatic pronunciation feedback. While the vision of a CBST that can replace human SLT (Speech and Language Therapist) is beyond the horizon of the next decade, recent advances has proved that computer can be a valuable “primary assistant” in the therapy process.