STARFISH: Safety and Speech Recognition with Artificial Intelligence in the Use of Air Traffic Control

Idiap will develop a (Deep Neural Network based) Automatic Speech Recognition (ASR) engine to operate over a tower phraseology in air-traffic management domain. We will also model typical grammar and phraseology deviations to deal with real-life scenarios. The developed ASR engine (i.e. related models such as acoustic and language models, dynamically updated dictionary) will be continuously updated and improved through the different iterations of the project. The focus of Idiap will be to deliver a V2T to a subsequent task allowing to extract relevant Air Traffic Control (ATC) concepts from the recognized word sequences. Idiap will integrate the Assistant Based Speech Recognition (ABSR) paradigm, i.e., relying on provided context information which mainly consists of predicted possible air traffic controller commands, dynamically updated.
Idiap Research Institute
Oct 31, 2020
Mar 31, 2022