Sequence-to-Sequence Speech Recognition for Air Traffic Control Communication
Keywords
Loading...
Authors
Issue Date
2020-07-12
Language
en
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
In the domain of air traffic control, systems that make use of automatic speech recognition could detect errors in communication, alleviate work load of air traffic controllers and extract information for air traffic management systems. These systems could improve the overall safety of aviation, possibly even saving lives in the process. Efforts to bring automatic speech recognition into the domain of air traffic control have been made as early as the 1990s. How-ever, the amount of research in this field remains limited. In particular, there is an absence of sequence-to-sequence models in this field. In this work I therefore set out to create auto-matic speech recognition models using sequence-to-sequence model architectures, and im-prove them for the domain of air traffic control. The model with the best performance attained a word error rate of 26.19% on noisy, low-quality audio data, whereas it attained a word error rate of 5.9% on clean audio data. These results indicate that a solid contribution to the field of automatic speech recognition for air traffic control has been made, and the absence of sequence-to-sequence models in this field has been concluded.
Description
Citation
Supervisor
Faculty
Faculteit der Sociale Wetenschappen