Data Augmentation for Arabic Speech Recognition Based on End-to-End Deep Learning

End-to-end deep learning approach has greatly enhanced the performance of speech recognition systems. With techniques, overfitting stills main problem with a little data. Data augmentation is suitable solution for problem, which adopted to improve quantity training data and enhance robustness models. In this paper, we investigate method enhancing Arabic automatic (ASR) based on end-to-end learning. applied original corpus increasing by applying noise adaptation, pitch-shifting, speed transformation. An CNN-LSTM attention-based encoder-decoder are included in building acoustic model decoding phase. This considered as state-of-art learning, best our knowledge, there no prior research employed ASR addition, language built using RNN-LM LSTM-LM methods. The Standard Single Speaker Corpus (SASSC) without diacritics used an corpus. Experimental results show that improved word error rate (WER) when compared same augmentation. achieved average reduction WER 4.55%.