Detecting incorrect reading of the Quran through deep talk

For the correct pronunciation of Arabic, it is necessary to formulate the correct sounds of each phoneme in all speech units. Reading the Quran correctly with proper grammar and tajweed rules can be challenging, especially for Arabic speakers. Therefore, those who wish to read the Quran accurately seek guidance from experts. In this study, a speech recognition and verification system was developed using end-to-end deep learning. The goal of this system is to create a Deep Speech Recognition (DSR) system that supports accurate Quran recitation for everyone, regardless of gender. Online accessible Quran recitations are typically recorded by male adult professionals. Consequently, an Automatic Speech Recognition (ASR) system trained on such data may not be sufficient for female or child readers. In this study, an attempt was made to address this gap by using a comparison dataset containing Quran recitations recorded by individuals of different ages and both genders. Using this dataset, a speaker-independent speech recognition (SR) system based on deep learning was developed, and its performance was evaluated using metrics such as Word Error Rate (WER) and Character Error Rate (CER). The aim is to demonstrate how an ASR system trained and optimized on data from one gender performs on data from the other gender. As the number of male readers in our dataset is greater than female readers, the system performs better in recognizing male voices. The model of the system was trained with voices recorded by individuals of different genders and ages, as well as the voices of 7 imams from the world's most famous imams, using a deep speech model, and the results yielded a WER of 0.098315 and a CER of 0.076785. These results indicate that the system operates successfully regardless of gender and age differences.