2016 Volume 43 Issue 3 Pages 116-125
This study tried to determine robust detection criteria for the croaker's pulse-structure sound, using field recordings in WAV, WMA, and MP3 formats. Acoustic analysis results identified four temporal parameters and one frequency parameter as sound features. These feature quantities depended on the season of the year, not the file format. We created an automatic filter for croaker-sound detection using fixed-width ranges of these five parameters, and automatically detected sounds from identical fish-sound files, or the same files with noise. We also examined the correct detection (CD) and false alarm (FA) rates of the automatic detection. Without any noise mixing, it showed fair performance, e.g., 0% FA. Once noise was mixed, however, the CD reduced in accordance with the signal-to-noise ratio. However, the FA level remained constant (no more than 4.5%).
