Geez Reading Level Classification by Audio Feature Extraction using CNN

The Ge'ez language is an ancient Semitic language of Ethiopian. This language originates from the region encompassing Ethiopian and Eritrea Orthodox Tewahedo Church. The language is given as a course in the Orthodox Tewahedo Church spiritual school. Ge'ez has three types of reading these are Ge'ez, wurid, and kume. Each type of reading can characterize by different features and become distinguishable with its unique feature. The proposed system has five components: data acquisition, preprocessing, segmentation, feature extraction, and classification. In Audio signal processing, we collected a dataset from a spiritual scholar with audio files and different spiritual websites. Data preprocessing transform the raw data into a useful and efficient format. Segmentation is the stage used to split the audio signal before changing into the spectrogram with equal time intervals. In feature extraction, we propose to apply a Gabor filter on the input spectrogram image for texture feature extraction. Finally, the proposed model classifies the input spectrogram image using the convolutional neural network approach for grading into a specific class (Ge'ez, Wurid, and Kume). The proposed system is implemented using Python Anaconda and tested using a sample spectrogram image dataset.