Towards an accurate speaker-independent Holy Quran acoustic model

The popularity of speech recognition tools keeps increasing with the processing power of mobile devices. The use of speech recognition for the Arabic in general and the Holy Quran, in particular, has also followed the same trend. Holy Quran speech recognition systems have been developed by increasing the training data. In this paper, a more accurate Carnegie Melon University Sphinx acoustic model was trained for the Holy Quran chapters 001, and 067 to 114. When more efforts were put into having a more accurate training data, the resulting Word Error Rate of trained acoustic model reached around 15%