Male and female red-cheeked cordon bleus sing similar yet individualistic songs

Abstract

Birdsong is an excellent system for studying complex vocal signaling in both males and females. Historically, most research in captivity has focused only on male song. This has left a gap in our understanding of the environmental, neuroendocrine, and mechanistic control of female song. Here, we report the overall acoustic features, repertoire, and stereotypy of both male and female Red-Cheeked Cordon Bleus (Uraeginthus bengalus) (RCCBs) songs in the lab. We found few sex differences in the acoustic structure, song repertoire, and song stereotypy of RCCBs. Both sexes had similar song entropy, peak frequency, and duration. Additionally, individuals of both sexes sang only a single song type each and had similar levels of song and syllable stereotypy. However, we did find that female RCCBs had higher song bandwidth but lower syllable repertoires. Finally, and most strikingly, we found highly individualistic songs in RCCBs. Each individual produced a stereotyped and unique song with no birds sharing song types and very few syllable types being shared between birds of either sex. We propose that RCCBs represent a promising species for future investigations of the acoustic sex differences in song in a lab environment, and also for understanding the evolutionary driving forces behind individualistic songs.

I. INTRODUCTION

Vocal communication is one of the primary modalities of social and sexual communication for many animals. Songbirds (oscine passerines) are a large avian taxon well known for their ability to produce complex and learned/experience-dependent vocalizations. The production and elaboration of birdsong has emerged as an important model for studying vocal communication, sexual/social selection, and complex signaling. However, the distribution of study has not been even across species and sexes within the songbirds (oscine passerines). The study of birdsong has historically considered male song in north temperate migratory birds (Riebel et al., 2005; Odom et al., 2014; Rose et al., 2022b). The situation has changed in recent years but there is still a substantial gap in our understanding of both female song and song in the tropics.

Temperate male bird song tends to be linked with short breeding seasons, rapid increases in testosterone, gonad growth, and intense inter and intra-sexual selection (Ball et al., 2020; Rose et al., 2022a; Rose et al., 2022b). In these species, female song (if present), tends to be shorter, less complex, and less stereotyped (Rose et al., 2022b). However, recent work has demonstrated that female song has a broader range of timing and functions than typical male song, and has questioned if its production is closely linked to seasonal state and testosterone in the same way that male song is thought to be linked to these two variables (Langmore, 1998; Slater and Mann, 2004; Rose et al., 2019). Additionally, female song in tropical species has been studied primarily in duetting species in which year-round territoriality and long-term pair bonds seem to be important factors in shaping patterns of song production (Slater and Mann, 2004; Odom et al., 2015). While this topic has been gaining attention, sex-differences in song behavior are still not fully understood. Sex-differences in tropical non-duetting species are especially understudied.

We have established a colony of red-cheeked cordon bleus (RCCBs) in the lab to study the song behavior and neuroendocrine control of female song in this tropical species. Red-cheeked cordon bleus (Uraeginthus bengalus) are small tropical birds native to sub-Saharan Africa. They are an Estrildid finch with two close sister species (the blue-capped and blue breasted cordon bleus). All of the cordon bleus live in pairs or small flocks on the plains of Africa and have both male and female song (Payne, 2020). The song and dancing behavior of the blue-capped cordon bleu has been studied in captivity (Ota et al., 2015, 2018). For example, recent work in the blue-capped cordon bleus found that while there were many sex-similarities in song structure, females had lower syllable repertoire sizes and near significant differences in song duration (Geberzahn and Gahr, 2011). However, relatively little is known about the RCCB. A recent study from our lab found that male and female RCCBs have no sex-difference in song production rate, circulating testosterone, and progesterone and also that they have one of the most similar sex ratios of song-control system volumes of any species (including the blue-capped cordon bleu) (Rose et al., 2023). We also found that male and female RCCBs had comparable levels of motor driven immediate early gene expression in HVC but not in the robust archipallium (RA) after singing.

Based on data presented in this recent study several questions arise: (1) do male and female RCCBs differ in gross song characteristics, (2) do male and female RCCBs differ in song or syllable repertoire size, and (3) do male and female RCCBs differ in song stereotypy? First, the results of our recent study suggest that male and female RCCBs may have similar song structure. Testosterone is a known factor contributing to song production in male songbirds and has also been shown to modulate song structure and stereotypy (Alward et al., 2018; Brenowitz and Lent, 2002; Ball et al., 2020). The lack of sex differences in circulating testosterone in this species suggests that they may have similar song structure. Additionally, it is likely that RCCBs have similar song structure given that this has been reported in their sister species (Geberzahn and Gahr, 2011). Second, the results of our recent study also suggest that male and female RCCBs may have similar repertoire sizes. The volumes of song control system nuclei have been correlated with repertoire size and overall song rate in some species (Devoogd et al., 1993; Garamszegi et al., 2005; Pfaff et al., 2007). Our findings that RCCBs have one of the smallest sex differences in song control system nuclei volumes suggest that they may also have highly similar song and syllable repertoires. Finally, our results showing the lack of motor-driven gene expression in the RA of male RCCBs raise questions regarding sex differences in song stereotypy (plasticity). Previous studies have reported that birds with increased vocal plasticity had higher motor driven gene expression in RA (Hayase et al., 2021). Therefore, it is possible that female RCCBs will have lower song stereotypy. However, as mentioned previously, our results showing similar levels of circulating testosterone in both sexes may indicate more similar song stereotypy in both sexes.

In this study, we followed up on our recent neuroendocrine findings in the RCCB by more thoroughly exploring sex-variation in song behavior. We measured overall song structure in male and female RCCBs. We also calculated song and syllable repertoires in both sexes and measured syllable production and syllable order stereotypy.

II. METHODS

A. Experimental animals

Eighteen adult red-cheeked cordon bleus (RCCBs) were bought from a local breeder (Maryland Exotic Birds) and placed on long days (13 L:11D) to induce breeding and a photostimulated state. These birds were kept in the lab several months before the beginning of the experiment. During this waiting period, birds were housed in mixed-sex groups of nine individuals in 46 × 91 × 51 cm cages. These cages were maintained in a colony room with both males and females present at the University of Maryland, College Park, MD, and all birds were provided food and water ad libitum. Birds were sexed by the presence or absence of red cheek plumage which is only present in males.

1. Experimental housing

After a few months of acclimation in the colony room, birds were placed in pairs (one male and one female) into isolated housing. Birds housed in this manner were given a nest, nesting material, and food and water ad libitum for two months or more. This two-month period facilitated pair bonding as seen by the increase in allopreening, dancing, and nest building. Three pairs also began egg laying during this time. All procedures were approved by the University of Maryland, College Park Animal Care and Use Committee.

B. Recording procedures

Both sexes of RCCBs can be most reliably induced to sing by separating a bonded pair (Rose et al., 2023). Therefore, to induce singing, we separated each pair one at a time. Birds were placed in a sound attenuated chamber (Industrial Acoustics Company, Bronx, NY, model IAC-3) which was lined with acoustic foam, and recorded for an hour using a Zoom F8 (Zoom, Hauppauge, NY) multitrack field recorder (sampling rate of 44.1 kHz). Over the course of several months, we recorded all subjects at least four times. All songs produced in all four recording sessions were assessed qualitatively. For each bird, we chose the recording with the most songs to analyze quantitatively.

We used the SongFinder matlab code (see supplementary material) to count and isolate the songs in each recording. The number of songs per recording ranged from 2 to 329 and averaged 93 songs. In support of previous findings (Rose et al., 2023) there were no significant differences in the number of songs performed by males and females. From the two male and two female birds who performed the most songs, we isolated a consecutive set of 150 songs and a second consecutive set of 25 songs. We analyzed the gross song measurements (detailed in the following) of each set of songs and found that the 25 song set captured all of the syllable types and the overall song variation of the 150 song sets. Therefore, for the remaining analysis, we included only 25 songs per individual.

C. Song level acoustic analysis

Gross song level measurements were taken in RAVEN Pro v1.6.1 (Center for Conservation Bioacoustics, 2019) using a 512 point discrete Fourier transform (DFT). Songs isolated by the SongFinder matlab code were opened in Raven Pro. For each song, we measured average entropy, peak frequency, 90% bandwidth, and 90% duration. These measurements are standard when using Raven software, which uses robust energy based metrics to estimate 90% of the energy to reduce the error from “by-eye” spectrogram measurements (Raven User's Manual, 2010) For each variable we then calculated mean and standard deviation. Finally, differences in song structure between male and female songs were compared using a two-tailed t-test.

D. Syllable level acoustic analysis

To analyze syllable level song variation, songs isolated by SongFinder were opened in a second matlab code SyllaBuster (see supplementary material). SyllaBuster was designed to streamline syllable classification and requires a parameter file from SongFinder to run. First, the computer makes automated recommendations for syllable divisions based on user input parameters. We estimated a difference of 10 dB between syllable peaks and inter-syllable gaps and parsed a signal bandpass filtered between 2000 and 20 000 Hz. This was largely effective, but the program does allow for hand adjustments of syllable divisions in the case of noise or other interference in the recording. Twelve acoustic variables are then automatically measured for each syllable and exported to Microsoft Excel. These include: average frequency, instantaneous frequency spread, peak frequency, 90% duration, 99% duration, 90% bandwidth, entropy, tone count, upward sweep rate, total sweep rate, temporal skewness, and temporal kurtosis (definitions in Table I). Afterwards syllables are presented one by one in matlab to allow for hand classification into syllable groups based on the displayed spectrogram. By spectrogram classification is a common method for syllable type classification (Sikora et al., 2021; Dos Santos et al., 2023). These same syllables can be reclassified by subsequent blind observers as long as the parameter file is saved in the same folder as the original .Wav files.

TABLE I.

Definitions of the 12 acoustic variables measured by SyllaBuster. Unless otherwise noted, "average" means weighted by the total power in each time or frequency step, as appropriate. ^*SyllaBuster does not track spectrogram continuity so, counterintuitively, the downward drop between two upward chirps may dominate this metric.

Variable name	Definition	Units
Average frequency	The weighted mean frequency from a power spectral density of the whole syllable	Hz
Instantaneous frequency spread	The frequency standard deviation of the spectrogram averaged over time steps. For example, a single-toned sweep would have high bandwidth across the whole syllable but low instantaneous frequency spread.	Hz
Peak frequency	The frequency of the single loudest tone occurring anywhere in the syllable. Measured by recording the maximum spectrogram element.	Hz
90% duration	The time between 5% and 95% cumulative power for the syllable, summed over all relevant frequencies	Seconds
99% duration	The time between 0.5% and 99.5% cumulative power for the syllable, summed over all relevant frequencies	Seconds
90% bandwidth	The difference between the frequencies of 5% and 95% cumulative power for the syllable, summed over all relevant times	Hz
Entropy	Average instantaneous entropy of the spectrogram. The time-average value of −∑i[Pi*ln(Pi)], where Pi is the normalized power spectral density vs frequency on the spectrogram. To properly scale the Tone Count (next metric), the smearing entropy of the Blackman window (1.02) was subtracted from each summation.	No units
Tone count	Average value of exp(−∑i[Pi*ln(Pi)]) using the definitions noted previously. For equal-powered tonal sounds, this calculation yields the average number of notes being sung simultaneously. This metric was included because it increases rapidly when a syllable contains atonal sounds.	No units
Upward sweep rate	Root-mean-square (RMS) rate of change of the average frequency signed so that upward sweeps are positive and downward sweeps are negative*	kHz/ms
Total sweep rate	RMS rate of change of the average frequency signed so that all sweeps are positive	kHz/ms
Temporal skewness	Normalized third moment of the power (total sound level summed over all relevant frequencies). Positive skewness means louder sound early in the syllable; negative skewness means louder sound at the end.	No units
Temporal kurtosis	Normalized fourth moment of the power (total sound level summed over all relevant frequencies). Always positive. Higher kurtosis means more sound at the start and end of the syllable.	No units

In this study, spectrograms of syllables from 25 songs per bird from nine male and nine female cordon bleus were classified independently by one experienced and two naive observers (Beecher et al., 1981; Byers, 1996; Terry et al., 2001). Observers were not given guidance on how many syllable categories to create. Before calculating accuracy, observers looked at their groups to determine if there were any obvious highly similar syllable groups that one observer split or combined differently than the others. A decision was made regarding lumping or splitting before category accuracy was calculated. After calculating accuracy, the three observers reassessed any disputed syllables and reclassified them.

Naive observers had an average agreement of 91.6% with the experienced observer before disputed syllables were reclassified. One male and one female had less than 75% agreement and a complete consensus on syllable grouping could not be reached. Therefore, these two birds were removed from any statistical analysis, but based on the groupings of the experienced observer the trends of these two birds are included in the discussion.

E. Statistical analysis

To assess stereotypy, we completed several analyses. First, we analyzed syllable sequence stereotypy. To do this, we compared syllable sequence between consecutive songs using the Levenshtein distance metric in R (Dos Santos et al., 2023). Levenshtein distance measures the variation between consecutive sequences by calculating the minimum number of edits required to convert one string of characters into another. Due to the large variation in song length between birds, we corrected the Levenshtein distance for each bird by dividing by the average number of syllables for each bird. This correction was necessary for scenarios in which a bird with 13 syllables changed two or three syllables per song transition, whereas a bird with five syllables might change one or two. In this scenario, the bird with longer songs would have a higher (less stereotyped) Levenshtein score despite changing a lower percentage of his or her syllables.

Next, we calculated syllable production stereotypy between songs of the same sequence from a single bird. We isolated five songs from each bird with the exact same sequence (e.g., ABCD, ABCD). We did not choose songs with any sequence variation to avoid changes in syllable production due to coarticulation (Wohlgemuth et al., 2010; Sonderegger and Yu, 2010; Beckers, 2011). Choosing songs with identical sequences will result in higher stereotypy values than comparing syllable production across all syllables of the same type. However, this method better represents a bird's ability to reproduce its own song and avoids any potential sex differences in how often a bird varies its sequence. For each syllable, we used the 12 acoustic variables measured from the Matlab code SyllaBuster. Within a bird, for each syllable type (e.g., A) we calculated the coefficient of variation for all 12 variables across the five songs. We then ran binomial generalized linear mixed models (glmm) in R in the package lme4. We set sex as the independent variable and individual ID as the random effect. We started with all 12 acoustic variables and went through variable selection based on the significance of the fixed effects and the Akaike information criterion (AIC). We also ran a null model including only 1+ random effects.

III. RESULTS

A. Gross song description and sex differences

Male and female RCCBs sing at similar rates (Rose et al., 2023). Each bird sang a single unique song that remained consistent across all four recording sessions (Fig. 1). No two birds sang the same song and few syllable types were shared between individuals. On average, birds shared 0.44 syllables between any two individuals. This low amount of sharing was similar between individuals of each sex (F–F: 0.39; M–M: 0.42; M–F: 0.47 syllables shared on average). These syllable types were only shared by up to five birds and were never isolated to a single sex. There were no sex specific songs or syllable types. Overall song structure in males and females (N = 9) did not significantly differ in average entropy (M = 3.97 ± 0.53 bits, F = 4.05 ± 0.63 bits; p = 0.76) duration (M = 1.29 ± 0.59 s, F = 0.88 ± 0.23 s; p = 0.052), or peak frequency (M = 6.58 ± 0.85 kHz, F = 5.93 ± 0.94 kHz; p = 0.17). However, females had significantly higher song bandwidths than males (p = 0.03; Fig. 2).

FIG. 1.

Sample songs from two individual female and two individual male red-cheeked cordon bleus. Each spectrogram shows two consecutive songs from the individual shown.

FIG. 2.

Sex differences in the average and standard deviation of gross acoustic features of red-cheeked cordon bleu song. Males and females did not significantly differ in average entropy, duration, or peak frequency. Females had significantly higher song bandwidths than males.

B. Syllable sequence stereotypy

Male and female RCCBs did not differ in syllable repertoire size (M = 11 ± 2 syllables, F = 8.4 ± 2.1 syllables, p = 0.18, N = 8). However, one female had a syllable repertoire of 19 syllables, which is double the next highest female. When we removed this statistical outlier, females had significantly lower syllable repertoire sizes (M = 11 ± 2 syllables, F = 7 ± 1.3 syllables, p = 0.001; Fig. 3). Additionally, although we did not find a significant sex difference in song duration (Fig. 2) the sexes did significantly differ in the number of syllables per song (M = 10.2 ± 3.6 syllables, F = 6.9 ± 1.8 syllables, p = 0.04). Finally, males and females did not differ in standardized Levenshtein distance (M = 0.33 ± 0.19 edits/syllable, F = 0.35 ± 0.22, p = 0.92; Fig. 3).

FIG. 3.

Male and female red-cheeked cordon bleus statistically differed in total repertoire size only after one statistical outlier female was removed (B). Individual data points are graphed to show the full distribution of each sex (A). Point size indicates the number of birds with that repertoire size. Mean and standard deviation are also shown. Red-cheeked cordon bleus did not differ in song sequence stereotypy measured as a corrected Levenshtein distance (C).

C. Syllable production stereotypy

Male and female RCCBS did not differ in syllable production stereotypy. One female and two males did not have at least five songs of identical sequence, primarily due to single syllable substitutions. Therefore, we included eight females and seven males in this analysis. We started with a model including ten acoustic variables (despite the overfitting due to sample size) so that we could assess the importance of each variable. Two variables (90% Duration and Upward Sweep Rate were removed for correlation to other variables). No variables were significant and we found no permutation of the full glmm that had a better AIC than the null model. Therefore, we report the p-values from the full model (Table II). The full model had an AIC of (46.58), the second best model contained only Tone Count (AIC = 25.4), and the null model had an AIC of (23.46). Therefore, we did not find that males and females differed in syllable production stereotypy by any of these acoustic variables.

TABLE II.

The average syllable coefficient of variation for each acoustic variable and the p-value for each variable resulting from the full glmm predicting sex. None of these acoustic variables were significantly different between the sexes.

Variable	Average F CoV	Average M CoV	p-value
Average frequency	0.08 ± 0.02	0.06 ± 0.02	0.861
Peak frequency	0.14 ± 0.04	0.10 ± 0.04	0.978
99% duration	0.13 ± 0.02	0.13 ± 0.05	0.999
90% bandwidth	0.32 ± 0.08	0.33 ± 0.07	0.892
Entropy	0.26 ± 0.07	0.27 ± 0.07	0.837
Skewness	−1.0 ± 2.5	3.7 ± 7.9	0.865
Temporal kurtosis	0.34 ± 0.10	0.33 ± 0.03	0.883
Tone count	0.25 ± 0.09	0.22 ± 0.10	0.787
Total sweep rate	0.45 ± 0.13	0.38 ± 0.09	0.902
Instantaneous frequency spread	0.30 ± 0.05	0.34 ± 0.12	0.824

IV. DISCUSSION

The most striking result of our analysis of RCCB songs is the high amount of inter-individual variation combined with a high degree of stereotypy within a single bird's song performance. Songbirds generally have a species-typical song in which there is some individual variation. However, we did not find a species-typical song in RCCBs. When in solitude, both male and female RCCBs produce a highly unique and individually identifiable song. Additionally, very few syllable types were shared between birds. Such an individually unique song is a rare occurrence. We have only found this reported in one other species, the blue-capped cordon bleu (Geberzahn and Gahr, 2011). These sister-species share a habitat as well as a similar social structure (Payne, 2020). Further investigation should explore the ecological or evolutionary factors that may have led to this divergence in song behavior from the other Estrildid finches.

The highly individual songs found in cordon bleus present a conundrum. Although they may point to the importance of signaling individual identity in this species, it is well known that this level of acoustic differentiation is not necessary for individual identification (e.g., Christie et al., 2004; Nelson and Poesel, 2007; Hahn et al., 2013). Another possibility is that the high variation in song structure may better assist local non-migratory individuals in assessing the singer's fitness via the learning accuracy of individuals to their tutors (Nowicki et al., 2002). A third option may be that what appear to be individual songs in the lab may be more akin to family line songs or highly localized dialects in the wild, which may help avoid inbreeding. More information is needed about the ecology, social structure and learning paradigm of male and female cordon bleus to assess how this highly unusual song pattern may benefit individuals of these species.

We also found a high degree of sex similarity in song gross structure and stereotypy. The only differences we found in song performance between sexes were that of overall bandwidth and syllable repertoire. Female RCCBs had on average higher song bandwidth but lower syllable repertoires. However, a single female had a much higher syllable repertoire of 19 syllable types (comparable to the average male). Additionally, the two birds removed from syllable analysis due to difficulty classifying syllable types (one male and one female) appeared to have larger than average syllable repertoires. Therefore, we suspect that across a larger sample of birds, males, and females may not differ in syllable repertoire size. This contrasts with findings in the blue-capped cordon bleu in which females had significantly smaller syllable repertoire sizes (Geberzahn and Gahr, 2011). This may be in part due to the higher degree of sex similarity found in the volumes of the song control system nuclei of RCCBs than blue-capped cordon bleus (Rose et al., 2023). However, there are many possible sources of variation that we do not understand. For example, the one female with a high syllable repertoire may have had an unusual learning environment or may have crystalized an unusual song.

The female we removed from the syllable analysis portion of this study had 15 syllable types according to the experienced observer. She was also included in a previous study (Rose et al., 2023). Songs recorded for the current study were taken just before tissue collection in our previous study. At that time, this female had one of the lowest circulating testosterone levels we recorded. Additionally, her ovaries and oviduct were undeveloped suggesting that she was not fully in breeding condition. Three other females and four males measured here were also included in that previous study. Across these seven birds, despite syllable repertoire and stereotypy being fairly consistent, they varied greatly in gonadal development and circulating hormone levels. These observations are in line with past work in the blue-capped cordon bleu and demonstrate that neither testosterone nor breeding state are positively correlated with song structure and stereotypy in this species (Geberzahn and Gahr, 2011). This contrasts with work in temperate breeding songbirds which report a strong modulating role of both testosterone and season in modifying song behavior (Bernard and Ball, 1997; Rose et al., 2022a; Rose et al., 2022b).

Finally, we did not find that males and females differed in syllable sequence stereotypy or in syllable production stereotypy by any of the 12 acoustic variables we measured. This is likely due to the high level of inter-individual variation (especially in variables such as skewness) overshadowing any differences between specific individuals of either sex. Previous studies have reported that individual differences in stereotypy may be reflected in motor-driven gene expression in RA (Hayase et al., 2021). Specifically, they reported that canaries with increased vocal plasticity between adjacent syllables had higher motor driven gene expression in RA. Previous work demonstrated that female RCCBs had greater motor driven ZENK expression in RA than males (Rose et al., 2023). However, we did not find that this was reflected by an increase in vocal plasticity in either syllable sequence or in syllable type production. Therefore, our findings here do not support the idea that the degree of motor driven gene expression in RA is entirely driven by vocal plasticity.

In conclusion, we report here that male and female RCCBs sing similar and highly individual songs. We found that their individually unique songs and high level of sex similarity are mostly in line with their sister species, the blue-capped cordon bleu. We found that the high degree of sex similarity in song production also reflects previous reports of highly similar nuclei volumes in the song control system. We believe that the cordon bleus represents a valuable system for further investigation into the function of individually unique songs, the learning basis for each individual's unique syllables and sequence, and the ecological or evolutionary forces that may have caused this divergence away from the typical song behavior of Estrildid finches and other songbirds.

SUPPLEMENTARY MATERIAL

See the supplementary material for matlab codes SongFinder and SyllaBuster for extracting and categorizing songs and syllables from longer recordings.

AUTHOR DECLARATIONS

Conflict of Interest

The authors have no conflicts of interest to declare. All methods were approved by the University of Maryland IACUC.

DATA AVAILABILITY

The data that support the findings of this study are available from the corresponding author upon reasonable request.