Phoneme-based automatic speech recognition for real-time Arabic command drone control

Phoneme-based automatic speech recognition for real-time Arabic command drone control

연구 분야: Artificial Intelligence

학회: Iran Journal of Computer Science

Real-time automatic speech recognition (ASR) systems are essential for advancing human–machine interaction in specialized fields such as drone control, where precision and resource efficiency are paramount. While deep learning models dominate ASR research due to their high accuracy, traditional stochastic models offer a robust alternative for addressing challenges associated with low-resource languages and constrained computational environments. This study presents a Modern Standard Arabic (MSA) speech recognition system based on a phoneme-based hidden Markov model (HMM) and Gaussian mixture model (GMM). Trained on a custom-developed corpus of Arabic drone commands, the proposed system achieves a word recognition accuracy of 93.9% using a 3-HMM, 16-GMM configuration. Performance evaluation on both laptop and embedded platforms demonstrates a maximum execution time of 0.089 s, memory usage under 4.6 MB, and an average real-time factor (RTF) of 0.082, confirming its feasibility for real-time deployment on cost-effective edge devices. In addition, a detailed syllabic and phonetic analysis reveals that heavy syllables combined with voiced alveolar consonants enhance phonetic distinctiveness and improve recognition accuracy. Comparative results indicate that the system delivers competitive performance with existing drone command recognition approaches while providing a reproducible framework for ASR development in other under-resourced languages and dialects.