FIGURE 2. Acoustic analyses.

a–c, Calculation of trill rate, frequency bandwidth and vocal deviation, illustrated for the C. parvulus song in Fig. 1. a, Oscillograms were used to measure trill rate, as the number of syllables produced per second (7.85 Hz here). b, Amplitude spectra were used to measure frequency bandwidth, the range of frequencies produced (1.97 kHz here), using a -24 dB amplitude cut-off criterion¹³ (dashed line). This criterion was chosen a priori to maximize the proportion of signal energy analysed while excluding background noise. Earlier analyses revealed that lower dB cut-off values (for example, -27 and -30 dB) regularly include background noise in spectral analyses, while higher cut-off values (for example, -21 and -18 dB) unnecessarily exclude signal energy. Amplitude spectra were computed at 32 kilopoints and smoothed to a frequency resolution of 300 Hz. c, For each trill type, average frequency bandwidth was plotted as a function of average trill rate (for example, filled square). Vocal deviations (dashed line) were calculated relative to an upper-bound regression (solid line)¹³, which was in turn calculated relative to the distribution of trill types across 34 emberizid songbird species (grey dots)¹³. d, Plot of all trill types for all species analysed, with minimum area convex polygons shown. Overlap in raw data among most species is apparent. G. magnirostris, green; G. fortis, dark blue; G. fuliginosa, pink; G. scandens, yellow; C. parvulus, light blue; C. psittacula , light green; C. pallida, red; C. olivacea, black.