Automatic Recognition of Arabic Poetry Meter Using Machine Learning, Template Matching, and Deep Learning

This paper discusses three methods to identify the Arabic poetry meter; namely: Machine Learning (ML), Template Matching (TM), and Deep Learning (DL). The Mel Frequency Cepstral Coefficients (MFCC) features were extracted. The MFCC features are used to represent the spectrogram and frequency domains, then they are used with ML, TM, and DL models. In TM, the following methods /algorithms are used: the Structural Similarity Index Measure (SSIM) and Oriented FAST, rotated BRIEF (ORB), and a type of Convolutional Neural Network (CNN)- which is the VGG16 deep learner. The final results show that ML outperforms other methods (TM, and DL) with an overall accuracy of 92.2%. The results are promising for auto recognition of the Arabic poetry meter -which is important for Arabic poetry learning.