Error Detection Using Syntactic Analysis for Air Traffic Speech

Abstract

Air traffic controllers and pilots communicate primarily through voice/speech to perform their day-to-day operations. Automatic speech recognition when applied in this domain can reduce the workload of both controller and pilot. The speech processing has several challenges in this domain like poor quality of radio channel, faster speaking rate and very strict vocabulary with infinite accent combinations. As errors have impact over the speech recognition process, the focus is on error detection in air traffic controllers speech when using automatic speech recognition. In this line, various error detection techniques available in normal English speech recognition are compared, and specific techniques which can help in air traffic controllers speech domain are discussed. Also, this paper emphasizes on syntactic analysis as a major component in the post-processing. Syntactic analysis along with phonetic string distance analysis helped to obtain close to 10% overall improvement in word error rate and 10–15% improvement in concept recognition rate for the experiments conducted over air traffic speech data considered for discussions.