Voice-Zikr: A Speech Recognition System Implementation for Hands-Free Zikr Based on Deep Learning

Speech recognition is a branch of pattern recog-nition that has been widely implemented in products. Some well-known products that used speech recognition systems include Google Assistant, Apple Siri, and Alexa which have high accuracy to produce output with user expectations. Recently, deep learning is one of the techniques that is often used to build models in speech recognition systems. The technique works to keep information in its hidden layers from audio frames as input features and phones as output labels respectively. Zikr is one of the Muslim worship activities that can be done at any time. Several tools and applications have been created to count the zikr words while repeatedly speaking them. In this research, the speech recognition system is implemented to create an application called voice-zikr that is used to count the zikr words spoken by Muslim people. The speech recognition model is trained using time delay neural networks with 5 hidden layers. The dataset was collected from different ages of speakers who read “Subhanallah”,” Alhamdulillah”, “Lailahaillallah”, and”Allahuakbar”. The model performance can reach 1.04 %WER on recorded audio testing and work perfectly on microphone testing,