Male songbirds sing to establish territories and to attract mates. However, increasing reports of singing in non-reproductive contexts and by females show that song use is more diverse than previously considered. Therefore, alternative functions of song, such as social cohesion and synchronisation of breeding, by and large were overlooked even in such well-studied species as the zebra finch (Taeniopygia guttata). In these social songbirds only the males sing and pairs breed synchronously in loose colonies following aseasonal rain events in their arid habitat. As males are not territorial, and pairs form long-term monogamous bonds early in life, conventional theory predicts that zebra finches should not sing much at all; yet they do and their song is the focus of hundreds of lab-based studies. We hypothesise that zebra finch song functions to maintain social cohesion and to synchronise breeding. Here we test this idea using data from five years of field studies, including observational transects, focal and year-round audio recordings, and a large-scale playback experiment. We show that zebra finches frequently sing while in groups, that breeding status influences song output at the nest and at aggregations, that they sing year-round, and that they predominantly sing when with their partner, suggesting that song remains important after pair formation. Our playback reveals that song actively features in social aggregations as it attracts conspecifics. Together, these results demonstrate that birdsong has important functions beyond territoriality and mate choice, illustrating its importance in coordination and cohesion of social units within larger societies.

This dataset contains data from several collection methods which together give an integrated perspective on wild zebra finch song ecology. The data was collected from 2016 to 2020 in UNSW Fowlers Gap Arid Zone Research Station in New South Wales, Australia, home to a well-studied population of wild zebra finches.

The methods are:

a) Nest box monitoring
We routinely checked nest boxes from 2016 to 2020, usually from August till December (in 2020 already from June onwards due to drought-breaking rain), when most breeding activity in this population is observed. We combined the resulting clutches laid with the rainfall data for our field site for 2016 to 2020 from the Australian Bureau of Meteorology.

b) Observational transects
In the spring of 2018, 2019 and 2020, we walked 1 km transects to standardise encounters with zebra finches at our field site. During these transects, for every zebra finch observation we scored the group size and presence of song.

c) Focal, hand-held acoustic recordings
In 2018, we opportunistically recorded wild zebra finch songs using directional microphones. Afterwards, using spectrograms made in Audacity, we scored for each recorded song which individual produced it and the specific social context in which it was produced.

d) Acoustic recordings at active nest boxes
In 2016, we installed time-programmable audio recorders near 63 active nest boxes, in various stages of breeding. We analysed one day (it was 26th October) from each nest using Audacity to score the start and end times of each singing bout, while also noting the quality (low, medium or high) as an indication of distance from the nest. We attributed each song to an individual by comparing spectrograms. We also assigned an owner for each nest based on song output and quality. For each nest, we had the breeding stage due to nest monitoring.

e) Year-round acoustic recording at activity hotspots
We placed time-programmable audio recorders (Song Meters 3 and 4, Wildlife Acoustics) that were active from sunrise to sunset every four days. For sound analysis, we selected the day of the month with the lowest average wind speeds per month for October 2018 – October 2019. We obtained these low wind days using a python script to link the recording timestamps to the half-hourly wind speed data from the local weather station (Australian Bureau of Meteorology) with the half hour closest to the start time of the recording used as indication of wind speed for that recording. Then, we calculated the average wind speed on the recording days and selected the day of the month with the lowest average wind speeds. We selected the four recorders on which we expected the highest chance of obtaining zebra finch activity, which were situated at feeders (3x) and an artificial water site (1x). Using sound spectrograms in Audacity we scored per 1-hour recording the presence of song and the vocal activity level, which considers that zebra finches regularly vocalise. This activity score was a 0 – 4 Likert-scale resembling the amount of vocal activity (0: no activity, 1: quiet overall, < 5 min of vocalisations, 2: not busy, 5 - 15 min, 3: quite busy, 15 - 30 min, 4: very busy, > 30 min). We also scored whether we heard any calls (less precise than the activity score) or zebra finches flying over (flying zebra finches do not sing).

f) Playback experiments which were filmed
We conducted a large-scale playback experiment from the 5th to 10th October 2017 that consisted of automated simultaneous playbacks with three treatments: zebra finch song, nightingale song (heterospecific control) and silence (additional control). Playback setups consisted of a nest box attached to a steel post next to a tree where a loudspeaker was placed. To control for local variation in zebra finch presence, treatments were organised in triplets (N = 23), with the three simultaneous treatments in the same general location. On a given day three or four triplet experiments were conducted at the same time in different areas of the study site, resulting in a total of 23 sites, of which 19 received visits by zebra finches. Played sound files consisted of two minutes of either zebra finch or nightingale song followed by 13 minutes of silence, resulting in a 15-minute loop playback. Playback experiments started between 06:45 and 10:05 in the morning and lasted about 7.5 hours. All treatments were filmed continuously and we scored the videos by noting the timestamp for all visits to the bush or nest box. We also scored foraging movements (birds only on the ground) or visits that happened after there were already birds present (these visitors might have been attracted by the birds instead of the playback). We also scored the sex and group size (pair or flock) of arriving birds. Additionally, we scored whether arriving males were singing although this was not always possible. We calculated for each arrival the visit latency, that is, how many seconds passed since the playback ended, scoring a zero if the visit was during playback.

For further details on how the data was collected, we refer to the (open access) related article.

All the uploaded datasets are csv files. The readme contains information on datasets and software, but the (open-source) software elements (R 4.1.1 analysis script; python 3.7.0 script linking wind speeds to audio files) are uploaded to Zenodo.