Web-based Arabic speech recognition system
Nurul Ashiekin Che Roslan · UMS Institutional Repository (Universiti Malaysia Sabah) · 2022
Speech recognition, also known as automated speech recognition (ASR), computer speech recognition, or speech-to-text, is a feature that allows a computer software to convert human speech into written text. ASR is a study of how computers can accept voice information from humans and translate it with the greatest probability of accuracy. Capturing and digitizing sound waves, translating them to simple language units or phonemes, creating vocabulary from phonemes, and contextually interpreting the word are all part of speech recognition. Speech recognition poses some interesting challenges such as varying acoustic conditions, dialects, and articulation at word's boundaries. Students may find it difficult to learn a new language, particularly the communication aspect, because pronunciation accuracy can be challenging. The suggested project is aimed to the students at University Malaysia Sabah who are currently studying Arabic language at the beginner level. The objectives of this project are to investigate the Arabic language speech recognition for beginner using Mel-Frequency Cepstrum Coefficient (MFCC) and Artificial Neural Network (ANN), to develop the web-based application for Arabic speech recognition using the Python and PHP and to evaluate the Arabic speech recognition and functionality of the web- based system. The speech samples that will be used in this project from the expert or someone can speak fluently in Arabic. The feature extraction, the Mel-Frequency Cepstrum Coefficient (MFCC) is used to improved detection accuracy. In this project, the implementation of Artificial Neural Network (ANN) is explored as a machine learning model in this project using a python. The project will be performed in waterfall model with the phases such as requirements, design, implementation, testing and operation. This project's result is occasionally unable to forecast the words uttered but still provides accuracy. Findings for this project is MFCC is that it is good in error reduction and able to produce a robust feature when the signal is affected by noise and ANN is a suitable classifier to use. However, a larger dataset is needed to get accurate prediction.