Research Article
Identification of Letters Hijaiyah Pronunciation Using Neural Network (Backpropagation) and Pre-Processing of Mel-Frequency Cepstral Coefficient
@INPROCEEDINGS{10.4108/eai.2-10-2018.2295283, author={Wafira Rahmania and Arini Arini and Anif Hanifa Setyaningrum and Arie Purnomosidi and Muhammad Taufik Rusydi}, title={Identification of Letters Hijaiyah Pronunciation Using Neural Network (Backpropagation) and Pre-Processing of Mel-Frequency Cepstral Coefficient}, proceedings={Proceedings of the 2nd International Conference on Quran and Hadith Studies Information Technology and Media in Conjunction with the 1st International Conference on Islam, Science and Technology, ICONQUHAS \& ICONIST, Bandung, October 2-4, 2018, Indonesia}, publisher={EAI}, proceedings_a={ICONQUHAS}, year={2020}, month={5}, keywords={signal processing; mel-frequency cepstral coefficient; artificial neural network (backpropagation); simulation;}, doi={10.4108/eai.2-10-2018.2295283} }
- Wafira Rahmania
Arini Arini
Anif Hanifa Setyaningrum
Arie Purnomosidi
Muhammad Taufik Rusydi
Year: 2020
Identification of Letters Hijaiyah Pronunciation Using Neural Network (Backpropagation) and Pre-Processing of Mel-Frequency Cepstral Coefficient
ICONQUHAS
EAI
DOI: 10.4108/eai.2-10-2018.2295283
Abstract
To avoid mistakes in pronouncing hijaiyah letters. The writer applies mel-frequency cepstral coefficient to extract and will yield characteristic value of voice signal. Implementation of Artificial Neural Networks (Backpropagation) is used for classification on the identification of 8 letters of hijaiyah using Matlab. 8 selected hijaiyah letters are س ص ذ ز ق ك ء ع take fathah. The feature extraction process produces several different parameter values, including pre-emphasis, windowing, fast fourier transform, discrete cosine transform, coefficient cepstrum and the duration. The backpopagation experiment using the maximum number of epoch and training functions varies as much as 15 times from each scenario capable of producing training regression 0.91019, test 0.93486, validation 0.99772 and MSE 0.2048. The test of hijaiyah pronunciation using trainlm with the number of hidden layer 10, obtained accuracy of 25%.