Arabic Speech Recognition Using Neural Networks

Abstract

Nowadays, speech recognition is essential in designing a natural voice interface for communication between human and their modern digital life equipment. But the obvious issue in this field is the lack of wide support for several universal languages and their dialects. This research comes to ensure the viability of designing the Automatic speech recognition model for the Modern Standard Arabic. The automatic speech recognition model was developed by training it to recognize each character of the Modern Standard Arabic . The model’s architecture followed the end-to-end speech recognition approach by using Mozilla DeepSpeech model which is a pretrained end-to-end neural network ASR in English language. The obtained result showed a word error rate as low as 24.3% and character error rate as low as 17.6%. Therefore, we concluded that the model can reach a much better word error rate by deploying any improvement such as combining more datasets. The applications of this research are vastly available such as developing a real-time speech recognizer for Arabic audio lectures.