AUTOMATIC SEARCH AND CLASSIFICATION OF SOUND SOURCES IN LONG-TERM SURVEILLANCE RECORDINGS
Robert C. Maher, Joseph G. Studniarz · 2012
The increasing availability of forensic audio surveillance recordings covering days or weeks of time makes human audition impractical. This paper describes the rationale and potential application of several techniques for high-speed automated search and classification of sound sources and sound events in long-term forensic audio recordings. Methods that can operate directly on perceptually compressed bitstreams without full decoding are of particular interest. Example applications include identification of aircraft overflights, the presence of human speech utterances, gunshot sounds, and other forensically relevant audio signals.