A Deep Learning Approach for Arabic Spoken Command Spotting

Keyword spotting (KWS) in spoken language involves identifying specific terms within an audio stream. While this task is widely employed in edge devices, achieving high accuracy is challenging due to the need for efficiency on low-power and resource-constrained systems. This paper presents a novel approach for Arabic keyword spotting using a ConformerGRU model architecture. The model's performance is further enhanced by training a text-to-speech model for synthetic data generation. The proposed method demonstrates state-of-the-art results with 99.59% accuracy.