Observational field studies provide insight on the multifunctional nature of birdsong. For example, if song production were limited to pre-fertilization, then that would suggest a mate attraction function. If it were used throughout the breeding season and in response to intruding males, then that would suggest a territorial defence function. In the present study, we determined the daily and seasonal singing patterns of male Ruby-crowned Kinglets (Regulus calendula) in Labrador, Canada, using microphone arrays in two breeding seasons. Using a playback experiment, we simulated a territorial intrusion to compare the structure of songs produced while defending a territory to the structure of songs produced during solo and contest singing. Singing peaked in the early part of the breeding season and then declined continuously for the remainder of the season, which suggests that the songs function in mate attraction. Singing peaked 2-3 h after dawn, and then declined steadily until it stopped at 2200 h. Some nocturnal singing was observed, but no dawn singing was observed. A high probability of signal overlap by heterospecific songs at dawn would hinder signal recognition and explain the observed delay in peak singing activity. Vocal responses to playback suggested a function in territory defence. However, there were no significant differences in the duty cycle, frequency modulation, and bandwidth of songs in relation to the context of song production, though songs were shorter in the intrusion context than during solo singing. Overall, the study provides the first quantitative description of the effects of time of day, time of year, and social context on singing behaviour in this understudied species.

Songs from the solo singing and counter-singing contexts were derived from the annotated songs from microphone array recordings. A song was considered 'solo' if it was not preceded or followed by the song of another male for 10 s (as determined by the time stamps of all annotated songs, not just those meeting the inclusion criteria), and 'countersinging' if it was. A song was considered 'intrusion' if it was produced by a focal male during the 5-min playback period or within one minute after the end of the playback period. We retained songs that were not masked by the playback for the ‘intrusion’ context. For ‘contest’ and ‘solo’ contexts, we retained only high-quality songs for our structural analysis, which included songs with no distortion or overlapping sounds, and that were produced within 15 m from an array microphone (as determined by the localization process). 

We measured the acoustic structure of each song using Luscinia software (http://rflachlan.github.io/Luscinia/). Each song was viewed as a spectrogram (512-point fast Fourier transform, Hamming window, 88% overlap, 28 dB dynamic range, 100% dynamic compression), and the automatic signal detection algorithm was used to identify the elements composing the song and the silent intervals between them. We measured the following four structural features from each song:

1. Song duration: the length of time between the beginning of the first element of the song and the end of the last, including the silent intervals between each element. The end of a song was determined when there were no elements produced within 4 seconds after the last element.
2. Duty cycle: the sum of the durations of each element divided by song duration.
3. Frequency modulation: the frequency of maximum amplitude (i.e. peak frequency) was determined for each spectrum within each element of the song. The cumulative absolute change in peak frequency from one spectrum to the next was then divided by the cumulative duration of all signal elements within the song to yield a rate of peak frequency change (Hz/s).
4. Bandwidth: the difference between the 95^th percentile and 5^th percentile of all peak frequency values in a song.

'element level data v02.csv' - contains the duration of each single element (i.e. a continuous note) in each song 
'song level data v02.csv' - contains information about the number of elements comprising a song as well as the duration of each song.
'peak frequency data v02.csv' - contains peak frequency measurements for the time bins compriseing each element in a song.
'summary stats from models.csv' - contains the mean and standard error of the song structure variables derived from the models run in the R script.

Fahmy Wilson RCKI song structure analysis R script:
This script calculates several structural variables from raw data created by Luscinia software (http://rflachlan.github.io/Luscinia/) and runs mixed effects models to determine the relation of the song structure variables to social context.
line 67 - used to relevel the treatments (i.e. social context) to obtain the mean and st error of the response variable from each model using the summary() function on the model result (e.g. line 73) and then document the mean and standard error of the intercept in an excel spreadsheat to create 'summary stats from models.csv', which is used to generate Figure 4. To relevel the dataframe,  change the 'ref =' to the different contexts (solo, context, intrusion) and rerun the models.