DEEP LEARNING UNTUK PENGENALAN PELAFALAN HURUF HIJAIYAH BERHARAKAT; DEEP LEARNING FOR RECOGNITION OF HIJAYAH LETTER UTTERANCE

RAJAGEDE, RIAN ADAM, Afiahayati

2016 | Skripsi | FMIPA

Development of software for Al-Qur'an learning has been developed in order to facilitate the users to learn the Al-Qur'an. However, there are problems in the process, one of the obstacles on the development of software-based Al-Quran is the difficulty of building a system that is able to recognize the pronunciation of an Arabic text by the user. Deep learning, an artificial neural network model that lately began bustling developed, has shown good results to improve the accuracy of voice recognition or other softcomputing cases. The principle in deep learning models is to make a deeper network to improve the accuracy of learning. Because of that, there needs to be some adjustments such as architecture, algorithm, and other optimizations which can support deep learning. In this research, deep learning model implemented for resolving cases of Arabic letter utterance recognition. Convolutional Neural Network (CNN) was used as architecture for solving the problem. Also some regularization optimizations were used to improve accuracy and reduce overfitting, such as dropout and L2 regularization. This research examined the sound recording and classified it into 10 classes Arabic letters. The results of this research, gained the accuracy 78.75% when performed without regularization and reached 80.75% when using regularization.

Kata Kunci : Deep learning, Convolutional Neural Network, Classification, Arabic letter