%Aigaion2 BibTeX export from Idiap Publications
%Thursday 04 December 2025 09:03:45 PM

@TECHREPORT{Piaget_Idiap-RR-04-2025,
                      author = {Piaget, Jehan Joachim Daniel and Prasad, Amrutha and Motlicek, Petr},
                    keywords = {Air-Traffic Communication (ATC), automatic speech recognition (ASR), end-to-end ASR, Prompting, Whisper models},
                    projects = {Idiap},
         mainresearchprogram = {Human-AI Teaming},
                       month = {7},
                       title = {Improving ASR and Callsign Detection in Air Traffic Control Speech using Whisper Prompting},
                        type = {Idiap-RR},
                      number = {Idiap-RR-04-2025},
                        year = {2025},
                 institution = {Idiap},
                     address = {Rue Marconi 19, Martigny, 1920},
                        note = {This semester project is done as a collaboration between EPFL and Idiap.},
                    abstract = {This report presents the semester project at EPFL, which was conducted in collaboration with the Idiap Research Institute. This work focuses on building a robust and reproducible pipeline to perform Automatic Speech Recognition (ASR) and callsigns detection in the context of Air Traffic Control (ATC) communications. Given the noisy, accented, and high-stakes nature of ATC speech, the goal is to explore and improve ASR performance with existing models.
The work involves evaluating different ASR models on the ATCO2 dataset. Experiments are conducted on
EPFL’s High Performance Computing (HPC) clusters using Apptainer containers, and investigate the use  of prompting techniques with the Whisper model to enhance transcription and callsign recognition accuracy.
The instructions and code to reproduce or extend the experiments can be found on the corresponding GitLab
repository.},
                         pdf = {https://publications.idiap.ch/attachments/reports/2025/Piaget_Idiap-RR-04-2025.pdf}
}