Consistent in recitation of the Quran is very important to become a perfect Muslim. The recitation has to be performed using correct rule of Tajweed in order to avoid recitation error that may leads to the mistranslation of the recited word or sentence. As the technology grown faster and so the number of Muslims, a variety of digital forms of Quran have been produced either in the form of mobile application or computer software. It is a good phenomenon as it will encourage more Muslims to recite the Quran. However, without the Tajweed knowledge, they tend to make mistakes during recitation. The traditional method of Quran Tajweed learning is based on Talaqqi and Musyafahah. This is a manual learning technique that involves face-to-face learning process between students and teacher. In order to address this issue, a Tajweed classification model based on digital speech processing technique and artificial neural network is developed as a fundamental research in this area. This study focuses on the rule of the Noon Sakinah and Tanween for the classification of Idgham with and without Ghunnah. The dataset is developed based on the Quran recitation from well-known reciters. Mel-Frequency Cepstral Coefficient is used for the feature extraction of the recitation sample. Meanwhile, the neural network is used for the Tajweed classifier. The training process of the neural network has been evaluated using three different training algorithms – Gradient Descent with Momentum, Resilient Backpropagation and Levenberg-Marquardt optimization. From the results, it can be concluded that the highest test accuracy is obtained by the Levernberg Marquardt training algorithm (77.7%) followed by the Gradient Descent with Momentum (76.7%) and Resilient Backpropagation (73.3%).