Cost Reductions Enabled by Machine Learning in ATM How can Automatic Speech Recognition enrich human operators performance?

Abstract

Various new solutions were recently implemented to replace paper flight strips through different means. Therefore, digital data comprising instructed air traffic controller (ATCO) commands can be used for various purposes. This paper summarizes recent works on developing speech recognition systems to automatically transcribe commands issued by airtraffic controllers to pilots allowing decrease of ATCOs’ workload, which leads to significant increase of ATM efficiency and cost savings. First experiments in AcListant® project have validated that Assistant Based Speech Recognition (ABSR) integrating a conventional speech recognizer with an assistant system can provide an adequate solution. The following EC H2020 funded MALORCA project has proposed new Machine Learning algorithms significantly reducing development and maintenance costs while exploiting new automatically transcribed speech corpora. In this paper, besides recapitulating achieved recognition performance for Prague and Vienna approach, new statistics obtained from various error analysis processes are presented. Results are detailed for different types of ATC commands followed by rationales causing the performance drops.