Robust Speech Recognition for Carrier Air Traffic Control

This proposal addresses the feasibility research on utilizing the speech recognition technology for carrier air traffic control. The speech recognition capability enables seamless integration of the Joint Unmanned Combat Air Systems (J-UCAS) with the aircraft carrier operations. To realize the speech recognition capability under this application scenario, we face two key challenges: the speech recognition has to work under the adversely noisy condition, and it has to meet the strict probability of success requirement. Recent progress in speech recognition technologies has spawned tremendous commercial applications. State-of-the-art automatic speech recognition systems are known to perform reasonably well under well-conditioned high signal-to-noise ratio condition. However, the system performance significantly degrades under real-world environments. Here, we would like to design a speech recognition system that is robust to non-stationary noise that is typical under the aircraft carrier operation environment. Various noise compensation methods will be evaluated and compared. In particular, we propose the use of missing-feature reconstruction method with the integration of the bone-conductive microphone. A preliminary design of the robust speech recognition system will be given under the Phase I research based on the comparisons of the system performance and computational complexity.BENEFITS: Robust speech recognition algorithms developed during the Phase I research will demonstrate the feasibility of deploying speech recognition technology for use in FATC. Phase II research will develop a prototype system that is adapted to the carrier air traffic control application. During the Phase III work, the prototype system will serve as the baseline system for integrating with the J-UCAS FATC architecture. The speech recognition prototype developed in Phase II research could be used in private sector ATC for the FAA. It could also be used in the Navy carrier ATC training simulation system. In addition to military applications, the robust speech recognition technologies developed in this research also benefit various commercial applications that require the recognition to work under adverse environments with the presence of non-stationary noise.