Quantifying the Benefits of Speech Recognition for an Air Traffic Management Application

Abstract

The project AcListant® (Active Listening Assistant), which uses automatic speech recognition to recognize the commands in air traffic controller to pilot communication, has achieved command recognition rates above 95%. These high rates were obtained with an Assistance-Based Speech Recognition (ABSR). An Arrival Manager (AMAN) cannot exactly predict the next actions of a controller, but it knows which commands are plausible in the current situation and which not. Therefore, the AMAN generates a set of possible commands every 20 seconds, which serves as context information for the speech recognizer.

Different validation trials have been performed with controllers from Düsseldorf, Frankfurt, Munich, Prague and Vienna in DLR’s air traffic simulator in Braunschweig from 2014 to 2015. Decision makers of air navigation providers (ANSPs) are primary not interested in high recognition rates, respectively, low error rates. They are interested in reducing costs and efforts. Therefore, the validation trials, that were performed at the end of 2015, aimed at quantifying the benefits of using speech recognition with respect to both efficiency and controller workload. The paper describes the experiments performed to show that with ABSR support controller workload for radar label maintenance could be reduced by a factor of three and that ABSR enables fuel savings of 50 to 65 liters per flight.