Challenges and Development of Speech Recognition in Air Traffic Control Applications
Researching a secure and reliable air traffic control(ATC)speech recognition system is of par-amount importance in advancing the automation process,reducing the workload of ATC personnel,and en-hancing safety efficiency.This paper provides a detailed overview of the challenges faced in the develop-ment of ATC speech recognition technology and evaluates commercial speech recognition systems from four companies:iFlytek,Tencent Cloud,Alibaba Cloud,and Shanghai Maitu.The results indicate that a-mong the three general-purpose commercial ASR systems,iFlytek achieves the best recognition perform-ance with a Character Error Rate(CER)of 25.36％.However,it falls short of meeting the deployment requirements in the ATC domain.In contrast,Maitu's product,which is trained on ATC data,demonstrates the optimal performance with a CER of 15.02％.Furthermore,this study designs experiments to explore the strengths and weaknesses of manually designed speech features and features extracted through self-supervised pre-training strategies.The findings show that the latter contributes to improved recognition accuracy and robustness but raises concerns about slower inference speeds and deployment complexities due to the large data volume.In conclusion,this paper summarizes its findings and provides a forward-looking perspective,enabling readers to gain a comprehensive understanding of the research in this field.Developing a safe and dependable ATC speech recognition system holds the key to propelling air traffic control into a more automated and efficient future while lightening the burden on controllers and enhan-cing overall safety.
万方数据
维普
