Study Species
The Ferruginous Antbird (Drymophyla ferruginea Temminck, 1822) is a species distributed in areas of dense Evergreen Forest of the Atlantic domain (Fig. 1A). It is found only in the Southeast of Brazil, mostly near coastal areas, but also extending inland from Minas Gerais and Sao Paulo (Fig. 1B). It inhabits the understory and mid-story of thickets in primary forests and also in second-growth woodland. It is an active forager that feeds on arthropods. This species is not globally threatened, and it is fairly common throughout its range.
File Survey
Vocalization recordings were obtained from Laboratory of Ornithology (School of Sciences, São Paulo State University), Xeno-Canto (www.xeno-canto.org), and Macauley Library (Cornell Lab of Ornithology, Cornell University). Each sound file represented a different individual. Files with excessive noise, poor recording quality, overlapping vocalizations, or any other factor that could hinder audio and spectrogram inspection were excluded from the sample. We sought to include recordings that cover as much as possible the species' geographic range that was proposed by. The coordinates and the year of recording were provided by the recordists of each vocalization. A total of 45 sound archives were sampled.
Vocalization Analysis
All the vocalizations were measured blind both to the identity and the location of the vocalization, to prevent the measurements from being biased in any way. The identity of the recording was extracted after the measurements. The digitized sound files were standardized (44.100 Hz, 16 bits of resolution, WAV format) using Sony SoundForge Pro 11 (Sony Creative Software, 2010). We randomly selected 3 vocalizations from each sound file, totaling 135 analyzed vocalizations. The vocalizations were selected according to the morphology of their syllables, which were visible by the spectrogram (Hanning Window, FFT = 512 points, Overlap = 93,75%) generated in the Sonic Visualizer software (Queen Mary, University of London, 2013). We employed the oscillogram to measure the time variables and the amplitude spectra (Hanning Window, resolution = 2048 points) to measure the frequencies (Fig. 1C and D). A limit of -42 db below the peak amplitude was taken on the power spectra to avoid the noise frequency bands. They were used to measure these variables because spectrograms have a trade-off between temporal and frequency resolution, which may lead to major problems on measurement. Both were generated using Sony SoundForge Pro 11 software and the variables were measured by the same person.
The following variables were obtained: T1- duration of the first syllable(s); T2- duration of the interval between the two syllables(s); T3- duration of the first frequency modulation (FM) of the second syllable(s); T4- duration of the second FM of the second syllable(s); T5- duration of the second syllable(s); T6- duration of the entire vocalization; F1- Maximum frequency of the first syllable (kHz); F2- Minimum frequency of the first syllable (kHz); F3- Frequency bandwidth (MAX-MIN) of the first syllable (kHz); F4- Peak frequency of the first syllable (kHz); F5- Maximum frequency of the second syllable (kHz); F6- Minimum frequency of the second syllable (kHz); F7- Frequency bandwidth (MAX-MIN) of the second syllable (kHz); F8- Peak frequency of the first FM of the second syllable (kHz); F9- Peak frequency of the second FM of the second syllable (kHz); F10- Difference between the peak frequencies of the second and the first FM of the second syllable (F9 minus F8) (kHz).
Statistical Analysis
We calculated the mean of the variables using three vocalizations for each sound file. We performed a principal component analysis (PCA) on the mean of variables to reduce its number, to avoid the colinearity of some variables and to decrease the data redundancy. We selected the components that had eigenvalues greater than|1.0|and extracted the scores of the principal component space for each individual. We created a "vocal distance" matrix among the individuals using the Euclidean distance between each pair of individuals in the principal component space. A matrix of "geographical distances" among individuals was created by Geographic Distance Matrix Generator 1.2.3 software (Center for Biodiversity and Conservation, American Museum of Natural History) employing the coordinates of latitude and longitude that were obtained from the sound files. We used the Mantel test to examine if the both matrices are correlated to each other. The factor "year of recording" was analyzed by a second Mantel's test. This test demonstrated that there was no effect of the year of recording on the characteristics of vocalizations. We also tested if the differences among the years of recording affects the vocal distances with another Mantel test. The "temporal distance" matrix were created using the Euclidean distance between the years of recording of each pair of individuals.