Assistant based speech recognition - another pair of eyes for the Arrival Manager

摘要

Nowadays Arrival Managers (AMAN) are available to produce efficient inbound traffic sequences and to create guidance advisories for optimized approaches. Information about deviations from the planned sequence is exchanged between controller and pilot via radio communication. The AMAN is only able to derive these deviations from the radar data. Using radar data as single input sensor, however, results in adaptation delays of 30 seconds and more - and even worse, the controllers' intent is still missing. The AcListant? AMAN (Active Listening Assistant) [1] has shown for the Dusseldorf Approach Area how to avoid this sensor delay by analyzing the controller-pilot-communication and using the gained information as an additional sensor. An Assistant Based Speech Recognition system (ABSR) is embedded in an AMAN, which provides a dynamic minimized world model to the speech recognizer. Validation trials were performed from February to March 2015 with seven male and four female air traffic controllers from Dusseldorf, Frankfurt, Munich, Vienna, and Prague. Depending on the accepted rejection rate of the speech recognizer, recognition rates between 90% and 95% were achieved, whereas without ABSR only rates between 58% and 83% were possible. Furthermore ABSR significantly reduces the deviation between the controllers' plan and the plan of the AMAN and, at the same time, significantly reduces the controllers workload.