DLR--Skyguide Cooperation Speech Understanding Results for Skyguide Voice Recordings: First Results of Step 3

Kurzfassung

From August to November 2024 DLR researchers and Skyguide have evaluated the performance of Automatic Speech Recognition and Understanding (ASRU) models on voice recordings of Skyguide. The models were originally trained and adapted with data from Austro Control for the Vienna airspace and had, therefore, never seen any data from airspaces controlled by Skyguide. The results were just based on 125 example utterances from air traffic controllers and pilots of Zurich enroute airspace. This report continues the evaluation, but is based on much more data. More than 4000 utterances from Zurich enroute, Zurich Approach and Zurich tower environment are considered. This report shall build a basis for adapting the ASRU system developed for Austro Control to the needs of Skyguide. Special focus is on the aspect of maintainability, which includes the recognition of new speakers and the addition of new waypoint and airlines names.