Population Differentiation in a Complex Bird Sound: A Comparison of Three Bioacoustical Analysis Procedures

Abstract

AbstractWe examined three bioacoustical analysis methods for comparing complex sounds among different populations. We chose the D‐syllable of the chick‐a‐dee call of the black‐capped chickadee (Poecile atricapilla) because it is a broadband sound representative of a class of vocalizations, common in many animals, that resists simple subjective classification for comparative studies. We examined the properties of the D‐syllable in field‐recorded samples from three different populations. The first method of data extraction sampled the amplitude values of a spectrum obtained in a single fast Fourier transform (SFFT) taken at the midpoint of each D‐syllable using multi‐speech software. The second method employed spectrogram cross‐correlation (SPCC) to obtain a matrix of similarity values between D‐syllables in the samples using canary software. The third method calculated similarity values obtained from the evaluation of four acoustic features of the D‐syllables derived from multi‐taper spectral analysis (MTSA) using sound analysis software. Following data extraction by these three techniques, we used multivariate statistical procedures to reduce the data for examination of differences among populations and to represent in scatter‐plots the patterns of clustering of the sounds. We found that the SFFT in the middle of the D‐syllable provided the poorest population discrimination following statistical processing, the SPCC method produced the next clearest population separation, and the MTSA method resulted in the most distinct separation of the three populations of D‐syllables. In carrying out these comparisons, we discovered that the characteristic environmental noise of a recording area can influence the signal properties of broadband sounds being compared by automated procedures, and could lead to faulty conclusions unless appropriate care is taken to mitigate the noise in which the signals of interest are embedded. Consequently we re‐analyzed our data following noise reduction and found less discrete population separation overall. However, the methods of SPCC and MTSA retained the ability to separate populations, with MTSA providing the sharpest discrimination among groups.