Genre Classification of Saint Yared Qum Zema Using a Convolutional Neural Network

Abstract Music Information Retrieval is one research area in Machine learning that emphasizes on extraction of valuable features from audio datasets for classification. The study shows the tasks of automatic genre classification of the greatest Composers of Saint Yared's Qum Zema, which is a form of Ethiopian sacred music sung in vocal voice. The three forms of zema genres are Geez, Ezil, and Araray, which have high intra-class similarity in classification. In this study, we have built a model called Saint Yared’s Qum Zema Classifier model to categorize audio data into three classes. The model took audio data and transformed it into spectrogram images. The total number of data used in this study were 1,555 audios, each audio segments in 10 equal seconds, and transformed into spectrogram images. The model uses ReLU activation function and Adam optimizer, which results a training accuracy of 98% and a testing accuracy of 88%. The model has the best result as compared with ResNet, VGGNet, and AlexNet in testing performance and model efficiency. The total average of the loss rate for the model was 1.01. This study provide base for further research in Ethiopian religious music and shows how deep machine learning can be used to preserve and classify intangible cultural treasures.