PROCESSING OF TRANSIENT SIGNALS

THE NAVAL UNDERWATER SYSTEMS CENTER HAS A CLEAR NEED FOR DETECTING AND CLASSIFYING ACOUSTIC SIGNALS. THESE SIGNALS MAY REPRESENT SHIPPING TRAFFIC, NATO NAVAL ACTIVITIES,, BIOLOGICALLY GENERATED SIGNALS (SUCH AS PRODUCED BY WHALES AND OTHER MARINE LIFE), ENVIRONMENTALLY GENERATED SIGNALS (CRACKING OF GLACIERS, ETC.), OR HOSTILE MILITARY ACTIVITY. SOME OF THESE SIGNAL SOURCES CAN BE MODELED FAIRLY WELL AS PRODUCING STATIONARY SIGNALS, AND THUS STANDARD FOURIER TRANSFORM METHODS CAN BE APPLIED SUCCESSFULLY. MANY SOURCES, HOWEVER, PRODUCE SIGNALS THAT ARE SHORT IN DURATION AND CHANGING WITH TIME, AND IT IS ON THESE SIGNALS THAT HELP IS NEEDED. WE PROPOSE TO AUTOMATE THIS DETECTION AND CLASSIFICATION OF TRANSIENT SIGNAL SOURCES BY EXTENDING AND IMPROVING METHODS THAT WE PREVIOUSLY HAVE APPLIED SUCCESSFULLY TO THE SPEECH AND SPEAKER RECOGNITION PROBLEMS. SPEECH IS A TRANSIENT SIGNAL; THE SAME SPEECH SOUND IS ALSO PRODUCED DIFFERENTLY BY DIFFERENT SPEAKERS. YET OUR METHODS ALLOW US TO IDENTIFY SOUNDS OR WORDS INDEPENDENTLY OF THE SPEAKER. WE BELIEVE THAT ACOUSTIC SOURCE IDENTIFICATION CAN BE LARGELY AUTOMATED BY USING THESE METHODS, WHICH GREW OUT OF FUNDAMENTAL RESEARCH ON INFORMATION THEORY AND ENTROPIC SIGNAL PROCESSING.