Quantifying singing behaviour
We generated a waveform and spectrogram (Hamming window, FFT = 512 samples, 87.5% overlap) for all 15-min recordings using Raven Pro sound analysis software (v1.5; Cornell Laboratory of Ornithology, Ithaca, NY). On each spectrogram, we drew cursor boxes (hereafter, “annotated”) around vireo songs that were visible on the spectrogram and waveform (i.e., clear pulses in amplitude). In some instances, songs from non-focal males could be seen and heard in the background of the recording, but these were easily distinguished from the subject’s songs either because they were relatively faint or because they did not match the known vocal repertoire of the subject. We defined songs as vocalizations comprising one or more elements, where elements of the same song are separated by < 0.5 s and those of different songs are separated by ≥ 0.5 s (Mejías et al. 2020, Mejías et al. 2021). Our song definition did not hinder our ability to identify discrete songs and rambling songs (Figure 2), as defined by Bradley (1980: Figures 1B, 2A, B, and C). In total, we annotated 17,682 vireo songs from 430 15-min recordings. To make our 15-min measure of vocal activity comparable to previous studies, we multiplied the number of discrete and rambling songs in each recording session by four to obtain hourly rates.