Quantifying complex patterns of bioacoustic variation: Use of a neural network to compare killer whale (Orcinus orca) dialects

Abstract

A quantitative measure of acoustic similarity is crucial to any study comparing vocalizations of different species, social groups, or individuals. The goal of this study was to develop a method of extracting frequency contours from recordings of pulsed vocalizations and to test a nonlinear index of acoustic similarity based on the error of an artificial neural network at classifying them. Since the performance of neural networks depends on the amount of consistent variation in the training data, this technique can be used to assess such variation from samples of acoustic signals. The frequency contour extraction and the neural network index were tested on samples of one call type shared by nine social groups of killer whales. For comparison, call similarity was judged by three human subjects in pairwise classification tasks. The results showed a significant correlation between the neural network index and the similarity ratings by the subjects. Both measures of acoustic similarity were significantly correlated with the groups' association patterns, indicating that both methods of quantifying acoustic similarity are biologically meaningful. An index based on neural network analysis therefore represents an objective and repeatable means of measuring acoustic similarity, and allows comparison of results across studies, species and time.

Diel and Spatial Dependence of Humpback Song and Non-Song Vocalizations in Fish Spawning Ground

“…Various approaches for vocalization classification have been applied to birds, land and marine mammals, including the application to acoustic censusing [62]. In cetaceans, various techniques have been employed in vocalization classification from their cepstral features including dynamic time warping [63], neural network [64], Gaussian mixture models, hidden Markov models [65], multi-class support vector machine model [66], and multivariate discriminant analysis [67]. Here we employ pitch-tracking to extract key features of humpback D-moan vocalizations and apply the centroid-based K-means [68] method to classify them.…”

Calling under pressure: short-finned pilot whales make social calls during deep foraging dives

“…and AO EFD , respectively) following Madsen et al [34]. Calls sampled at 192 kHz were decimated by a factor of 2, after which a spectrogram was computed with 5 ms Hanning windows (480 samples, zero-padded to 4096 samples for fast Fourier transform (FFT) computation) with 50 per cent overlap for a spectral resolution of 200 Hz and a temporal resolution of 2.5 ms. A supervised trace of the fundamental frequency contour [35] was used to derive the fundamental minimum (F min ), mean (F mean ) and maximum (F max ) frequency over the 95 per cent energy window. Spectral power distribution was estimated using the Welch method [36] by summing power spectra within the 95 per cent energy window.…”

Acoustic identification of nine delphinid species in the eastern tropical Pacific Ocean

“…The use of alternative classification methods, such as artificial neural networks, may be another way to increase the accuracy of whistle classification. Artificial neural networks operate in a non-linear, self-organizing way and therefore may be able to detect differences among species that would be missed by other statistical methods (Deecke et al 1999). Artificial neural networks have been successfully utilized to recognize the calls of bowhead whales (Potter et al 1994) and to measure the similarity of discrete calls of killer whales (Deecke et a/.…”