AcListant - System Requirements for Communication between AMAN and speech recognizer

Abstract

The project AcListant® will show for Düsseldorf airport and its airspace, how automatic speech recog-nition (ASR) and an Arrival Manager system (AMAN) may benefit from each other. The Arrival Manag-er gets an additional input sensor. ASR receives context information to significantly reduce the word and concept error rate. This report details the interface description between ASR and AMAN. Concrete syntax, types and values are defined for the communication between these two components. In addition, the AMAN’s reaction on possible ASR input and the integration of both components into the validation environment are explained. Section 2 describes the possible XML tags (Extensible Markup Language) of the commands of the speech recognizer, its syntax and possible values. The commands, sent by the speech recognizer (ASR), are not all directly used by the AMAN “4D-CARMA” (Four Dimensional Cooperative Arrival Manager). The speech interface module (ModSpeechTCP_Interface) uses its context information to combine some commands (section 3) into the same command (e.g. HEADING to TURN_LEFT_HEADING, if the current heading is known). Other commands are split into more de-tailed commands (e.g. a TURN_RIGHT_HEADING command into TURN_RIGHT_BASE, if the down-wind segment is left). We detail this in section 4. The transformed commands are stored into 4D-CARMA data base and then used by the different modules. In section 5 we describe the semantics of each command with respect to the following actions of 4D-CARMA. Section 6 shows some strategies of detecting errors in speech recognition and what should be done in the AMAN then. The integration of the two components AMAN and ASR “Kaldi” is explained in section 7.