Flight operations, being conducted both in commercial and military sectors, demand reliable and real-time radio communication between the aircraft and air traffic controlling (ATC) agencies. Undesirable background and channel noises during radio transmission (RT) increase pilots’ listening fatigue and pose a potential of serious air safety hazard. Established speech enhancement methods such as spectral subtraction (SS) are crucial in enhancing the quality of communication by separating speech from background noise in the transmitted audio signals along with the basic requirement of reliability and low latency. This paper proposes a speech enhancement technique for real-time audio communication over radio between pilots and ATC based on SS and a Deep Neural Network (DNN). Spectral Subtraction estimates noise in the audio spectrum and subtracts it from the noisy speech signal. The DNN is trained using a dataset of clean and noisy speech signals recorded in real-world aviation scenarios. Signal synthesis module is utilized to calculate magnitude and phase of the speech; where the magnitude is modified by the neural network, and then it is reconstructed with the original phase. The proposed methods are evaluated using objective measures such as signal-to-noise ratio (SNR), perceptual evaluation of speech quality (PESQ), as well as subjective evaluations by human listeners in the form of mean opinion score (MOS).
Speech Enhancement of Real-Time Radio Communication Between Aircraft and Air Traffic Control Based on Spectral Subtraction and Deep Learning
International Conference on Aeronautical Sciences, Engineering and Technology ; 2023 ; Muscat, Oman October 03, 2023 - October 05, 2023
2023-12-26
8 pages
Article/Chapter (Book)
Electronic Resource
English
- ON-DEVICE REAL-TIME TRAFFIC SIGNAL CONTROL SYSTEM BASED ON DEEP LEARNING
European Patent Office | 2022
|- ON-DEVICE REAL-TIME TRAFFIC SIGNAL CONTROL SYSTEM BASED ON DEEP LEARNING
European Patent Office | 2023