LAUSR

Being the main spiritual source and reference for Muslims, The Holy Qur’ān can be recited in ten recitations (Qiraat). Each recitation (Qiraah) possesses certain features and characteristics that can be discriminated using Tajweed rules, which can best be defined as the elocution rules for reciting The Holy Qur’ān. This paper describes our efforts towards preparing, designing, developing, and evaluating a large-vocabulary speaker-independent and continuous speech recognizer for The Holy Qur’ān based on the narration of Hafs from A’asim by utilizing the state-of-the-art Automatic Speech Recognition (ASR) evolutionary approaches. Several Tajweed rules as depicted from the narration of Hafs from A’asim have been addressed and embedded in the development of the speech recognizer in our work. In addition, this paper presents the preparation process of The Holy Qur’ān speech corpus, which was used to train and test the speech recognizer. For training the acoustic model in our speech recognizer, four experimental setups were used within KALDI toolkit that are different in terms of dataset size and Tajweed rules. The best experimental setup is based on Time Delay Neural Networks (TDNN) with sub-sampling technique and obtained a Word Error Rate (WER) in the range of (0.27–6.31%) and a Sentence Error Rate (SER) in the range of (0.4–17.39%). Therefore, the experimental results are very promising and they indicate that the speech recognizer is able to recognize The Holy Qur’ān based on the narration of Hafs from A’asim.