(Massachusetts Institute of Technology and Woods Hole Oceanographic Institution, 2009-02)
Severson, Jared
Marine mammal whistle calls present an attractive medium for covert underwater
communications. High quality models of the whistle calls are needed in order to synthesize
natural-sounding whistles with embedded information. Since the whistle calls
are composed of frequency modulated harmonic tones, they are best modeled as a
weighted superposition of harmonically related sinusoids. Previous research with bottlenose
dolphin whistle calls has produced synthetic whistles that sound too “clean”
for use in a covert communications system. Due to the sensitivity of the human auditory
system, watermarking schemes that slightly modify the fundamental frequency
contour have good potential for producing natural-sounding whistles embedded with
retrievable watermarks. Structured total least squares is used with linear prediction
analysis to track the time-varying fundamental frequency and harmonic amplitude
contours throughout a whistle call. Simulation and experimental results demonstrate
the capability to accurately model bottlenose dolphin whistle calls and retrieve embedded
information from watermarked synthetic whistle calls. Different fundamental
frequency watermarking schemes are proposed based on their ability to produce natural
sounding synthetic whistles and yield suitable watermark detection and retrieval.