Matching a frequency contour (pulse-repetition rate as a function of time) of a pulsed call of transient killer whales (solid line) to a reference contour (dotted line) using standardization of call length [panel (a)] and dynamic time warping [local extension and compression of the time axis of the frequency contour to maximize frequency overlap—panel (b)]. The match (given as the average similarity in frequency in percent for all points of the two contours) is 69.9% using standardization, but 86.9% using dynamic time warping.
Categorization of frequency contours of bottlenose dolphin whistles using an ART2 neural network and dynamic time warping to calculate similarity. Numbers represent individual whistle contours. Signature whistles are shown in bold and boxes identify signature whistles from the same individual. Signature whistle categories that were split by the analysis are linked with dotted lines. See Janik (1999) for visual representations of the whistle contours.
Effect of the vigilance on the categorization of 50 frequency contours from calls of transient killer whales. Panel (a) shows the increase in the number of categories generated with increasing vigilance. Panel (b) shows the change in the variance ratio (ratio of within- to between-category variance) with increasing vigilance and panel (c) shows the change in the variance ratio with increasing numbers of categories. This ratio reached a maximum at a vigilance of 81.24% (10 categories). Trend lines in panels (b) and (c) are sixth-order polynomials.
Results of the categorization of frequency contours from 20 randomly chosen calls of transient killer whales. All frequency contours in the same column were assigned to the same call type by the analysis. The reference contours representing each category are shown in the first row. Labels give the recording session (in the format yy-mm-dd) for each frequency contour.
Article metrics loading...
Full text loading...