Masterarbeit
Detecting Callsigns in Air Traffic Control Voice Communications
Completion
2025/12
Research Area
Students
Sejal Sameer Deshpande
Advisers
Abubaker Gaber
Prof. Dr.-Ing. Martin Gaedke
Description
In aviation, air traffic management plays a crucial role. Air traffic controllers (ATCos) and flight crews (FCs) need to communicate important information like commands, confirmation of reception of commands and other information related to the air traffic environment. These commands are communicated in the form of callsigns. While prior research work involved use of acoustic and traditional models for speech recognition and callsign extraction which required use of huge process pipelines and extraction of features from audio data at different stages which caused unnecessary complications and high consumption of resources. Recent research work also involved use of an end to end model for automatic speech recognition using Deep Speech. There are many situational risks involved in callsign detection, including cross cultural communication variations, background noise, misidentification due to callsign similarity, overlapping conversations, possibility of missing important callsign. This thesis focuses on using a new approach for recognizing callsigns using a composite framework.
To achieve this, modern automatic speech recognition algorithms, natural language processing (NLP) and large language model (LLM) are required to be implemented at particular stages. The thesis answers questions like how to automatically identify callsigns, evaluating the advantages of using NLP and LLM at different stages for detecting callsigns. It aims to check the effectiveness of NLP for processing the transcription outcome and how using LLM could enhance the detection by addressing missed callsigns, thereby evaluating the overall reliability of the proposed approach if used in air traffic management. The objective of this master thesis is to find an approach or a combination of approaches to solve the previously mentioned problem in the context of air traffic management. This particularly includes the state of the art methods for callsign detection in air traffic communications including: i. Speech to text conversion using ASR model. ii. NLP for processing the transcription output. iii. LLM for post-processing or contextual enhancement. Lastly, results will be studied comparatively for analyzing the feasibility or applicability of the proposed approach.
