Abstract:
Language is one of the essential aspects of human behavior, which has a major role in our daily life activities. It helps in transferring information and knowledge for generations in the form of written and speech format. Natural Language Processing (NLP) is used to perform useful tasks involving human language using an electronic device, tasks like enabling and improving human-machine, human-human communication, or simply doing useful processing of text or speech. Pronunciation detection is a technique used in speech processing for identifying the pronunciation style from speech signals.
Nowadays computer-assisted language learning (CALL) system can provide many possible benefits to both the language learner and teacher by processing text and speech data. They allow the continuous reaction to the student without requiring the individual helpfulness of the teacher or scholars, however in the area of pronunciation training, which often requires the full attention of the teacher for only single learners, therefore the aim of this study to design pronunciation detection. Whereas Gəʾəz language used the religious language in Ethiopia, specifically in Ethiopian Orthodox Church at this time. And also it came to a research area in different fields. In Ethiopia, different kinds of literature for several ancient manuscripts, arts (q∂ne), scriptures, and heritages, historical, ethical and religious chronicles are written by Gəʾəz language. Those literature, books, and any manuscripts contain different thought and attitudes on the philosophy, creativity, knowledge, civilization, ethical and cultural principle of Ethiopia. The usability of those books like the bible and any pray of books are a day to day activates of the people. The main problem to understand that literature and books are the difficulties of Gəʾəz language pronunciation. Mispronunciation is lead to distortion of the meaning and alter of POS. So, this thesis is focused to design a model for solving such a type of problems and challenges.
The acoustic features are used for pronunciation detection using MATLAB with ANN learning algorithm. The chosen acoustic features of speech signals could be easily represented into the system and training for the target pronunciation. After suitable times of training process, extra test signals are load into the system for pronunciation detection, which gives an accuracy of 76.9% classification rate.The result shows that the selected acoustic features (such as speech rate or duration, energy, pitch, formant and MFCC) are proven to be good representations of Gəʾəz language pronunciation for the speech signal. Finally, this thesis has comprehended the design automatic pronunciation detection for Gəʾəz language.