InterSpeech 2021

Detecting English Speech in the Air Traffic Control Voice Communication
(Oral presentation)

Igor Szöke (Brno University of Technology, Czech Republic), Santosh Kesiraju (Brno University of Technology, Czech Republic), Ondřej Novotný (Brno University of Technology, Czech Republic), Martin Kocour (Brno University of Technology, Czech Republic), Karel Veselý (Brno University of Technology, Czech Republic), Jan Černocký (Brno University of Technology, Czech Republic)
Developing in-cockpit voice enabled applications require a real-world dataset with labels and annotations. We launched a community platform for collecting the Air-Traffic Control (ATC) speech, world-wide in the ATCO² project. Filtering out non-English speech is one of the main components in the data processing pipeline. The proposed English Language Detection (ELD) system is based on the embeddings from Bayesian subspace multinomial model. It is trained on the word confusion network from an ASR system. It is robust, easy to train, and light weighted. We achieved 0.0439 equal-error-rate (EER), a 50% relative reduction as compared to the state-of-the-art acoustic ELD system based on x-vectors, in the in-domain scenario. Further, we achieved an EER of 0.1352, a 33% relative reduction as compared to the acoustic ELD, in the unseen language (out-of-domain) condition. We plan to publish the evaluation dataset from the ATCO² project.