Bridging the Gap: Developing an Automatic Speech Recognition System for Egyptian Dialect Integration Into Chatbots

Chatbots offer a dynamic and intelligent conversational experience, especially the ones that comprehend speech. Despite their utility, a significant gap remains in Arabic, particularly in Egyptian speech chatbots. One reason is the lack of Automatic Speech Recognition (ASR) systems that transcribe Egyptian speech. Millions of users use the Egyptian accent, underscoring a substantial demand for relevant ASR technologies. Therefore, the work presented in this paper introduces the development of an ASR system by implementing the Conformer and Wav2Vec 2.0 architectures, marking their first application in the Egyptian dialect. The system was trained on the EACSC dataset. Various ASR models were evaluated, and the leading model shows an advancement over established Arabic and Egyptian speech recognition models, achieving state-of-the-art performance on the EACSC dataset. Additionally, a chatbot utilizing the ASR model prototype is also presented.