Assistant-Based Speech Recognition for ATM Applications

In Proceedings of the 11th USA/Europe Air Traffic Management Research and Development Seminar (ATM 2015).


Situation awareness of today’s automation relies so far on sensor information, data bases and the information delivered by the operator using an appropriate user interface. Listening to the conversation of people is not addressed until today, but an asset in many working situations of teams. This paper shows that automatic speech recognition (ASR) integrating into air traffic management applications is an upcoming technology and is ready for use now.

Apple’s Siri® or Google’s Voice Search® are based on hundreds of thousands of hours of training data. This paper presents an assistant based speech recognition system (ABSR), based on only 40 hours of training data. ABSR uses speech recognition embedded in a controller assistant system, which provides a dynamic minimized world model to the speech recognizer. ASR and assistant system improve each other. On the one hand, the latter significantly reduces the search space of the first one, resulting in low command recognition error rates. On the other hand, the assistant system gains benefits from ASR, if the controllers’ mental model and the model of the system deviate from each other. Then the controller cannot rely on the output of the system anymore, i.e. the assistant system is useless during these time intervals. By using ABSR the duration of these time intervals is reduced by a factor of two.

Marc Schulder
Research Associate in Computational Linguistics

My research interests include natural language processing, sign languages and sentiment analysis.