Automatic Speech Recognition and Understanding Over Noisy Air Traffic Control VHF Channels in Singapore

In recent years, the demand for Automatic Speech Recognition (ASR) and Spoken Language Understanding (SLU) systems within the Air Traffic Control (ATC) domain has been increasing, especially systems that can be applied in practice. These systems are essential for reducing the workload of pilots and air traffic control officers (ATCOs) and ensuring the utmost accuracy in communication between pilots and ATCOs. However, ATC remains a low-resource and challenging domain. This paper presents our work on developing an ASR engine and an SLU system for ATC in Singapore, addressing these challenges. We introduce the Singapore Air Traffic Control (S-ATC) dataset, aimed at fostering research in this demanding field. We then discuss our contributions to constructing an efficient ASR system tailored for the ATC domain. Experimental results are provided to evaluate the effectiveness of combining an ASR system with a Natural Language Processing (NLP) model versus an End-to-End system for the SLU tasks in this specific domain. Additionally, we try to implement a model for Speaker Role Detection (SRD) task and propose ideas to enhance the efficiency of these systems in the ATC domain in the future.