Automatic Detection of Alarm Sounds in Cockpit Voice Recordings

Abstract

Analysis of non-speech contents in Cockpit Voice Recorder (CVR) is getting more and more attentions today. Here the author proposed an automatic detection and recognition algorithm, which deal with non-speech alarm sounds in CVR recordings with high efficiency and reliability. It employed spectrum features and time-domain morphological features as template components, classified sound frames based on template matching, and filtered the rough results to meet the rationality. In an experiment enrolling 25 randomly selected CVR recordings, efficiency of the proposed algorithm was indicated with high accuracy.