Pengelasan Sebutan Huruf Hijaiyah menggunakan Teknik Pembelajaran Mesin (Classification of Hijaiyah Letters Pronunciation using Machine Learning Techniques)

Mel-frequency cepstral coefficients (MFCC) features and classification techniques based on machine learning are often used in classifying hijaiyah letter pronunciations, however, the classification accuracy performance of hijaiyah letter pronunciations is still low even with the use of machine learning algorithms and MFCC features. Therefore, this study to analyze the features and relevant machine learning techniques will be presented in this study paper. In addition, the number of hijaiyah letters was also increased to 30 letters following the Uthmani resm. This research aims to prove that the suitable feature and relevant classification technique allows for precise classification of the pronunciation of each letter even with large amounts of letters. This research is conducted based on the six main stages in research methodologies which includes signal processing, feature searching, processing and feature selection, classification and lastly, testing, evaluation and analysis. The sampling rate used for all speech signal processing modules in this study is 44.1 kHz. The findings of the study show that the MFCC feature is the most suitable feature to classify the pronunciation of hijayah letters compared to other features that have been extracted based on the rank in the feature selection results. Comparison of accuracy performance shows that Random Forest (RF) classification technique achieves high accuracy by using MFCC feature, which is an average of 97 similar to 99% for each hijaiyah letter compared to other classification techniques that have been tested in this study. In conclusion, the use of MFCC features and RF classification techniques are able to provide a high performance of hijaiyah pronunciation classification accuracy, which is 98.29% on average even with the use of 30 letters.