Shafee’una Al-Qur’an

 

Title of the Project

Shafee’una Al-Qur’an

Students Details

201811712 Tala Taha Al Ahmed 

201610236 Jamila Ali Al-Ali 

Abstract

Shafee'una Al-Qur'an web application is one of the websites that can benefit from the recent breakthroughs in Artificial Intelligence (AI) research to help the user in reading, reciting, and memorizing the Holy Qur’an. The website can recognize what the user is reciting, detect errors in the recitations, and alert the user to such errors with vibration or audio/visual cues. This enables the user to memorize the Holy Qur'an and review it efficiently and effectively without requiring assistance from another person. Users of the Shafee'una Al- Qur'an web application can also search for verses of the Noble Qur’an directly through recitation or voice search, a feature that works very efficiently. The most distinctive feature of our immediate memorizer is that the user may learn any Surah or Ayah from the Quran depending on his needs. This feature allows the user to learn at his own pace while keeping track of his progress. At the core of the website is an effective Automatic Speech Recognition (ASR) system that can work on Quranic recitations. We started working on the most recent and exciting model for ASR, which is called Whisper. This model achieves human-like accuracy levels in recognizing speech in dozens of languages including Arabic. The source of this model is OpenAI, which is behind some of the most amazing models such as GPT-3 and DALL-E [2]. Several recordings by different people were used to test the model’s accuracy and its ability to detect errors. We notice that the model failed to transcribe some of the test cases with more than 35 words taking more than 10 minutes without converging. For the cases that the model completed, two were transcribed perfectly with Word Error Rate (WER) of 100%, while two exhibited some errors with WER 67%. These results are very encouraging compared to what is reported in a recent study about customizing DeepSpeech to recognize recitations by males/females with varying levels of proficiencies.