A Comparative Study of Speaker Role Identification in Air Traffic Communication Using Deep Learning Approaches

Automatic spoken instruction understanding (SIU) of the controller-pilot conversations in air traffic control (ATC) requires not only recognizing words and semantics speech but also determining role speaker. However, few published works on automatic systems communication focus speaker identification (SRI). In this article, we formulate SRI task as a binary classification problem. Furthermore, text-based, speech-based, speech-and-text-based multi-modal methods are proposed to achieve comprehensive comparison task. To ablate impacts comparative approaches, various advanced neural network architectures applied optimize implementation text-based speech-based methods. Most importantly, (MMSRINet) is designed by considering both textual modality features. aggregate features, modal fusion module fuse squeeze acoustic representations attention mechanism self-attention pooling layer, respectively. Finally, approaches validated ATCSpeech corpus collected from real-world ATC environment. The experimental results demonstrate that all worked for task, MMSRINet shows competitive performance robustness compared with other seen unseen data, achieving 98.56% 98.08% accuracy,