Arabic speech recognition by end-to-end, modular systems and human

Recent advances in automatic speech recognition (ASR) have achieved accuracy levels comparable to human transcribers, which led researchers debate if the machine has reached performance. Previous work focused on English language and modular hidden Markov model-deep neural network (HMM–DNN) systems. In this paper, we perform a comprehensive benchmarking for end-to-end transformer ASR, HMM–DNN (HSR) Arabic its dialects. For HSR, evaluate linguist performance lay-native speaker new dataset collected as part of study. ASR 12.5%, 27.5% , 33.8% WER; milestone MGB2, MGB3, MGB5 challenges respectively. Our results suggest that is still considerably better than with an absolute WER gap 3.5% average.