Reducing Controller Workload by Automatic Speech Recognition Assisted Radar Label Maintenance

Abstract

Various new hard- and software centered methods were recently implemented to replace paper flight strips through modern technical solutions. These solutions provide valuable information for other ATM applications about the verbally given guidance instructions from air traffic controllers (ATCos), but also tend to increase ATCo workload. Speech recognition applications, which automatically recognize and input the verbal ATCo instructions into a technical flight strip solution, can compensate the workload increase while maintaining the benefit for other ATM applications. Experiments from the AcListant® project in 2015 have shown that Assistant Based Speech Recognition (ABSR), which combines a conventional speech recognizer with an assis-tant system, can provide adequate recognition quality for the use in air traffic control (ATC) applications. AcListant® used a prototypic radar display implementation and a research proto-type of a speech recognizer. This paper describes the exercise 220 of the SESAR 2020 funded solution PJ.16-04. It used a Commer-cial-Off-The-Shelf (COTS) speech recognition engine instead of a research prototype. Furthermore, a radar display developed by Thales Air Sys served for visualization. Command recognition rates varied greatly between 31% and 82% for different control-lers. However, the concept from ABSR to predict possible ATCo instructions could be integrated with the COTS engine, which significantly decreased the command recognition error rate and led to a variation between only 4.8% and 6.6%, i.e. only a small amount of false recognitions were shown to the ATCo.