Evolution of species recognition when ecology and sexual selection favor signal stasis

These data were generated over an 11-year period in a study of song recognition among four related bird species in the Himalaya (Cyanoderma ambiguum, C. ruficeps, C. chrysaeum, C. pyrrhops), as well as close relatives. The data primarily consists of measurements of songs and notes of all species, and the strength of responses to playback of songs, some of which were experimentally manipulated for frequency.

Methods
Abstract These data were generated over an 11-year period in a study of song recognition among four related bird species in the Himalaya (Cyanoderma ambiguum, C. ruficeps, C. chrysaeum, C. pyrrhops), as well as close relatives. The data primarily consists of measurements of songs and notes of all species, and the strength of responses to playback of songs, some of which were experimentally manipulated for frequency.

Methods During the breeding seasons 2011-2021, we visited localities across the Himalaya between one and ten times. We recorded songs using Telinga Twin Science microphones with a Sony PCM-D50 solid state recorder, or later with a Sound Devices Mixpre3 digital audio tape recorder. All the recordings were made in .wav file format with 16-bit resolution and 44.1-kHz sampling rate, or later with 24-bit resolution and 48 kHz sampling rate. For each recording, we noted altitude, date and locality. For the experimental analysis of frequency, we manipulated songs using the “Change Pitch” function in Audacity (https://audacityteam.org/). Tapes for playback were made from clean recordings, with any noisy portions removed if needed. Amplitudes were standardized to similar levels, just below the maximum possible. We located a singing male and conducted a playback for 5 minutes. We noted responses on a four-point scale, based on distance from speaker and directional movement (we also recorded flyovers, which correlate with the metric used, and are available on request). Some males were tested sequentially with different tapes, usually the playing of conspecific song after a trial with no response to the first tape, but on other occasions three or four tapes were also played, with a maximum of 4 sequential trials (to 10 males).

From the recordings, we compiled four sets of measurements. First, we measured characteristics of entire songs. For each song, we measured minimum and maximum frequency, center frequency (the frequency that divides the selected sound spectrum into two frequency intervals of equal energy), bandwidth 90%—the difference in frequencies separating the spectrogram into 5% and 95% energy quantiles—and peak frequency (the frequency with highest energy), with a Hann Window of size of 1024 samples. We measured song length, with a window size of 256 samples. We also counted the number of notes and number of distinct notes in each song. We scored songs for whether they contained an introductory note (or on occasion two introductory notes) which we define as at least 50% greater space between this note (or pair of notes) than the other notes. Second, taking a single clear song from each bout, we measured all notes in the song for frequency parameters, note length and qualitatively recorded note shape, as upsweep, downsweep, dome, saucer or flat, according to whether frequency monotonically increased, monotonically decreased, showed an intermediate peak, an intermediate minimum, or did not change. We split each note into four equal time intervals and recorded the center frequency of each interval. We then took the deviation of each center frequency from the average of the four frequencies (for example for a flat note, the deviations would be 0,0,0,0). We extracted principal components from the correlation matrix derived from the 4 columns of deviations x 1921 rows (notes). Correlations of the first principal component with the four center frequency residuals are: 0.94, 0.41, -0.84, -0.9, implying notes with large positive scores are downsweeps, and those with large negative scores upsweeps. Finally, we computed a measure of amplitude from the waveform, dividing a single note into 0.0002 second intervals, and computing the root mean square of amplitudes within each interval.