Penerapan Metode Mel Frequency Cepstral Coefficients dan Convolutional Neural Network untuk Pengenalan Bacaan Al-Qur’an sebagai Bagian dari Automatic Speech Recognition

Abstract

How to read the Qur'an correctly is one of the important issues in Islamic education. The purpose of this study is to develop a model to evaluate the correctness of the Qur'an reading using datasets obtained from Kaggle. The stages of the model development include preprocessing (trimming and denoising), feature extraction process using Mel Frequency Cepstral Coefficients (MFCC), and building a classification model using the Convolution Neural Network (CNN) method. The Mel Frequency Cepstral Coefficient (MFCC) is used to extract relevant features from speech samples. The process of extracting speech characteristics is carried out by applying the stages of the Mel Frequency Cepstrum Coefficient. The model was tested in four scenarios with variations in MFCC coefficients as well as the use of trimming and denoising. The test results show that the use of MFCC 128 coefficient without trimming and denoising provides the highest accuracy of 77%, while the use of MFCC 13 coefficient without trimming and denoising provides the lowest accuracy of 68%. This outcome is expected to be applied in applicatios (apps)  that can assist  people learn to read the Qur'an.