AUTOMATIC SPEECH RECOGNITION FOR AFAAN OROMO: A DEEP LEARNING APPROACH

Major Advisor: Jabesa Daba (Assist. Professor) Speech is one of the common styles of communication among human beings. Since human being‘s desire to communicate with machines using speech, developing an automatic speech recognition system was an interesting research area. According to the finding of the research conducted by Kasahun Gelana, the performance gained for Afaan Oromoo language is highly promising and as the language is becoming one of the most spoken language of the country, conducting a research on Automatic Speech Recognition will have an enormous advantage for development of the language. Thus, in this study, we used a deep learning approach to develop a limited vocabulary speaker independent speech recognition system for Afaan Oromo. The research work is implemented using a recurrent neural network (RNN) as a model along with different development tools to get the desired result. In order to achieve the objective of this research work, a total of 5100 speech sentences corpus is collected and prepared in a format suitable for use in the development process and classified as training and test set. We have used 80/20 rule for training and testing the model. The model was trained on 5.01 hrs of voiced data and its corresponding transcription. The method was evaluated on 7 Minutes of testing data. The performance of the trained model on the test set was good, given that the data was devoid of any background noise and lack of variability. The performance of the model is measured using LSTM Model. The study has saved meaningfully more keystrokes and the test has shown a result in keystroke savings of 0.87%. Keywords: - Afaan Oromo Speech recognition, Recurrent Neural Network, Long short Term Memory.