Accession Number : ADA321975

Title :   Development and Performance of a Cockpit Control System Operated by Voice: Summary Report of Project DMKLu/ACO2/A/9lO5.

Descriptive Note : Final rept.,

Corporate Author : HUMAN FACTORS RESEARCH INST TNO SOESTERBERG (NETHERLANDS)

Personal Author(s) : Steeneken, H. J.

PDF Url : ADA321975

Report Date : 29 NOV 1996

Pagination or Media Count : 16

Abstract : Phase 3 of this project is concerned with the evaluation of an automatic speech recognizer for cockpit control functions in the MLU-F16. The report is a summary of the following 3 reports: (1) Automatic speech recognition performance in a simulation-based fast-jet cockpit application, (2) Spontaneous-speech data base for cockpit control applications applied to commercial state-of-the-art speech recognition technology, and (3) Evaluation of integrated automatic speech recognition on the WSF mid-life update F-16 simulator. A total of 29 sessions were flown during shake-down and training stages yielding 32.5 hours of recording. In 17 of these sorties three RWLAF pilots were participating. The overall achieved word recognition accuracy was around 0.69, with scores per session ranging 0.53 to 0.88. The average completion rate (i.e. correctly executed commands) was around 66%. This is a low performance and insufficient for the envisioned operational applications. The pilot debriefing information learned that although the performance was considered insufficient, the expansion of functions, such as radio station selection by name, was highly appreciated. In general the present syntax was too complex which lead to incorrect commands. Also the awareness of the node status of the recognizer was marginal. A more flexible command language is an important requirement. With the recorded speech signals of 17 sorties a data base was compiled. With this data base a repetition of the recognition experiments can be made with different types of recognizers. Assessment of a new large vocabulary speech recognizer which was trained for the grammar (command string construction) of the cockpit commands produced a significantly higher recognition performance (0.87).

Descriptors :   *SPEECH RECOGNITION, *FLIGHT SIMULATORS, *WORD RECOGNITION, DATA BASES, ADAPTIVE CONTROL SYSTEMS, FLIGHT CONTROL SYSTEMS, COCKPITS, NETHERLANDS, PILOTS, HUMAN FACTORS ENGINEERING, JET FIGHTERS, WORDS(LANGUAGE), VOCABULARY, FLIGHT SIMULATION, SYNTAX, VOICE COMMUNICATIONS, REACTION CONTROL SYSTEMS, FLIGHT RECORDERS.

Subject Categories : Flight Control and Instrumentation
      Voice Communications
      Attack and Fighter Aircraft
      Human Factors Engineering & Man Machine System

Distribution Statement : APPROVED FOR PUBLIC RELEASE