TU Delft / AE / CO / CNS-ATM / Master thesis assignment:
ATC radiotelephony voice communication data mining
Current air traffic control commands between air traffic controllers (ATCo) and pilots are primarily handled through radiotelephony (RT) voice communication. At the same time, many studies have envisioned a future system where communication is digitized, or even automated.
Platforms such as LiveATC provide access to these real-time and recorded communications. This data provides us a source for deriving information from these voice communications, and thus, provides the potential knowledge for future automation.
Several research projects in the past had focused on constructing speech recognition methods to automatic transcribe ATCo voice commands. However, few leads to an open model that can be used freely. This thesis project aims at constructing an open reference dataset and a functional neural network model for ATCo speech recognition, which would be used for future ATM human-in-the-loop studies.
This project aims at combining open-source speech recognition libraries, neural networks, open air traffic data to generate an open model that can convert ATCo voice to text. Potentially, the following topics will be explored:
- Identify the challenges and advantages of speech recognition in air traffic management.
- Construct training datasets by labeling audio communication with text-format procedural commands.
- Design neural networks (CNN and/or RNN) that can be trained specifically for ATC communications.
- Incorporate trajectory information from ADS-B and Enhance Mode-S to improve model accuracy.
- Study the accuracy and uncertainty of the model
Analyze output data and study the potential benefits for air traffic management research.
- Setup an open framework allowing future efforts to improve the model.
In addition to related aeronautics background knowledge, we are looking for an MSc student who:
- is interested in cross-domain research (ATM, human-machine, and computer science);
- has a high level of programming skills (Python);
- shows strong interests in machine learning and neural networks;
- is willing to dedicate effort for manual labeling of training dataset;
- can think and manage the progress independent, as well as collaboratively with the supervisors;
- have a strong sense of time management.
- Dr. Junzi Sun (firstname.lastname@example.org) [Daily supervisor]
- Prof.dr.ir Jacco Hoekstra (email@example.com)
- Helmke, H, et al., Increased acceptance of controller assistance by automatic speech recognition. In Tenth USA/Europe Air Traffic Management Research and Development Seminar, 2013.
- Nguyen, V. N., & Holone, H. Possibilities, challenges and the state of the art of automatic speech recognition in air traffic control. International Journal of Computer, Electrical, Automation, Control and Information Engineering, 9(8), 1940-1949, 2015.
- Kleinert, M., et al., Machine learning of controller command prediction models from recorded radar data and controller speech utterance, in 7th SESAR Innovation Days, Belgrade, 2017.
- Kleinert, M., et al., Semi-supervised Adaptation of Assistant Based Speech Recognition Models for different Approach Areas, in 2018 IEEE/AIAA 37th Digital Avionics Systems Conference (DASC) (pp. 1-10). IEEE, 2018.
- Pellegrini, T., et al., The Airbus air traffic control speech recognition 2018 challenge: towards ATC automatic transcription and call sign detection, 2018.
- Chen, S. et al., Characterizing National Airspace System Operations Using Automated Voice Data Processing, ATM Seminar, 2019
© Copyright 2020 Junzi Sun