An Automated Procedure for Evaluating Song Imitation

Songbirds have emerged as an excellent model system to understand the neural basis of vocal and motor learning. Like humans, songbirds learn to imitate the vocalizations of their parents or other conspecific “tutors.” Young songbirds learn by comparing their own vocalizations to the memory of their tutor song, slowly improving until over the course of several weeks they can achieve an excellent imitation of the tutor. Because of the slow progression of vocal learning, and the large amounts of singing generated, automated algorithms for quantifying vocal imitation have become increasingly important for studying the mechanisms underlying this process. However, methodologies for quantifying song imitation are complicated by the highly variable songs of either juvenile birds or those that learn poorly because of experimental manipulations. Here we present a method for the evaluation of song imitation that incorporates two innovations: First, an automated procedure for selecting pupil song segments, and, second, a new algorithm, implemented in Matlab, for computing both song acoustic and sequence similarity. We tested our procedure using zebra finch song and determined a set of acoustic features for which the algorithm optimally differentiates between similar and non-similar songs.     

Imitation and Invention in Song Learning in Nightingales (Luscinia megarhynchos B., Turdidae)

Song learning in oscine birds is often defined solely as a process of song imitation; nonetheless, not all songs produced by laboratory‐tutored birds are imitations of the model songs. If song learning were strictly a process of imitation, these non‐imitated songs (inventions) would be expected to contain no learned attributes. To determine whether species‐typical song attributes can be learned in the absence of imitation, we compared the imitations and inventions of laboratory‐tutored nightingales (Luscinia megarhynchos B.) with the songs of wild nightingales and the songs of laboratory‐reared, untutored nightingales. The species‐typical song attribute measured was stereotypy. We quantified stereotypy by four variables: (1) percentage of notes shared between two renditions of the same song type (2) difference in repetition rates of the same trill in two renditions of the same song type (3) acoustic similarity of the same note in two renditions of the same song type, and (4) acoustic similarity of the same note repeated within a trill. Wild songs and imitated songs were significantly more stereotypical than the songs of untutored birds for all measures. For the percentage of notes shared (1), and the acoustic similarity of notes in two renditions of the same song type (3), invented songs did not differ from the songs of untutored birds, suggesting that imitation is necessary for the acquisition of these song characteristics. However, invented songs were significantly more stereotypical than the songs of untutored birds for measures of stereotypy related to trills (2 and 4), and neither imitated nor invented songs differed significantly from the songs of wild birds in terms of trill rate stereotypy (2). Thus, it appears that the process of learning to produce trills may differ from the process of learning non‐repetitive song components: increased stereotypy in trills occurs even when the trills themselves are not copied from song models. Strict imitation does not fully account for the acquisition of some learned song attributes.     

Non‐Salient Geographic Variation in Birdsong in a Species That Learns by Improvisation

Geographic variation in birdsong and differential responses of territorial males to local and non‐local song variants have been documented in a number of songbird species in which males learn their songs through imitation. Here, we investigated geographic song variation and responses to local and non‐local song in the grasshopper sparrow (Ammodramus savannarum), a species in which males develop song by improvisation rather than imitation, as a first step toward understanding how the extent and salience of geographic song variation is related to the mode of song development. To describe the geographic variation in song, we compared songs from populations in eastern Maryland and central Ohio, USA, using multiple acoustic analysis techniques. We then conducted a playback experiment in Maryland using local and non‐local (Ohio) songs to test how territorial males responded to this geographic variation. We found acoustic differences between songs from the two sites. However, males responded similarly to playback of these songs, suggesting that this geographic variation is not behaviorally salient in a territorial context. Together with previous studies, our results suggest that across species, geographic song variation and the extent to which this variation functions in communication may be correlated with the accuracy with which song models are imitated during song development.     

Community convergence in bird song

Species in similar habitats are often similar in morphology or behaviour, attributed to adaptation to similar environmental selection pressures, sometimes mediated by competitive interactions. For passerine songs, similarity of phenotype in identical habitats and character displacement have been documented, the former due to adaptation to the acoustics of the habitat, and the latter due to competition for acoustic space among species. If these phenomena are widespread, they should lead to community convergence of bird songs. Here, we test if passerine communities in similar habitats converge in song attributes or in acoustic differentiation among species. We compared the songs of European and North American Mediterranean climate passerine communities in open and closed habitats. Song frequency varied across different habitats but not continents. This was independent of both phylogeny and body size, indicating community convergence due to acoustic adaptation, rather than species sorting or similarity as a by-product of another type of ecological convergence. We found little evidence for regular spacing in song features among species, as would be expected if acoustic competition shapes within-community structure. However, for one of five song components, the open habitat communities showed a similar distribution of phenotypes on each continent. The proportion of interspecific variation in song explained by these effects was small. The fact that songs are complex signals that vary in many dimensions may explain why competition for acoustic space seems to be of small importance in structuring songs in these passerine communities.     

Sexually dimorphic phrase organization in the song of the indris (Indri indri)

Animal acoustic communication often takes the form of complex sequences, composed of multiple distinct acoustic units, which can vary in their degree of stereotypy. Studies of sequence variation may contribute to our understanding of the structural flexibility of primates' songs, which can provide essential ecological and behavioral information about variability at the individual, population, and specific level and provide insights into the mechanisms and drivers responsible for the evolutionary change of communicative traits. Several methods have been used for investigating different levels of structural information and sequence similarity in acoustic displays. We studied intra and interindividual variation in the song structuring of a singing primate, the indri (Indri indri), which inhabits the montane rain forests of Madagascar. Indri groups emit duets and choruses in which they combine long notes, short single units, and phrases consisting of a variable number of units (from two to six) with slightly descending frequency. Males' and females' contributions to the song differ in the temporal and frequency structure of song units and repertoire size. We calculated the similarity of phrase organization across different individual contributions using the Levenshtein distance, a logic distance that expressed the minimum cost to convert a sequence into another and can measure differences between two sequences of data. We then analyzed the degree of similarity within and between individuals and found that: (a) the phrase structure of songs varied between reproductive males and females: female structuring of the song showed a higher number of phrases if compared to males; (b) male contributions to the song were overall more similar to those of other males than were female contributions to the song of other females; (c) male contributions were more stereotyped than female contributions, which showed greater individual flexibility. The picture emerging from phrase combinatorics in the indris is in agreement with previous findings of rhythmic features and song repertoire size of the indris, which also suggested that female songs are potentially less stereotyped than those of males.     

Differentiation of Mating Vocalizations in Birds: Acoustic Features in Mainland and Island Populations and Evidence of Habitat-Dependent Selection on Songs

Local environments can act as selective agents on some characteristics of birds’ songs, whereas other song traits may not reflect local genetic adaptation. Geographic variation in songs of two Australian bird species (red‐capped robins Petroica goodenovii, western gerygones Gerygone fusca) was studied to examine one component of the ‘habitat‐dependent selection’ hypothesis. This hypothesis suggests that: (1) the detailed spectral characteristics of male songs are an evolved response to local habitat conditions affecting signal propagation and detection and (2) parallel evolution of other fitness traits sets up the potential for assortative mating by female choice. To examine the first part of the hypothesis, I made comparisons among widespread mainland populations and an island population using two levels of analysis: a typological analysis of song morphology (phonology: notes, syllables, syntax, temporal pattern, repertoires) and a spectral analysis of acoustic characteristics of songs (mean frequency, Wiener entropy, frequency modulation) using an automated procedure of feature extraction (Sound Analysis Pro). Spectral analysis was also used to extract values of the same acoustic features from the background sound environment of each recorded population. The typological analysis revealed no differences among mainland populations of either species, but large differences between mainland songs and those on the island. In contrast, the spectral analysis revealed acoustic divergence among populations, both mainland and island. For both species, Wiener entropy of songs correlated negatively with that of the ambient sound environment, consistent with predictions of the habitat‐dependent selection hypothesis of environmental selection on signal design.     

Song parameters of the fuscous honeyeater Lichenostomus fuscus correlate with habitat characteristics in fragmented landscapes

Both avian abundance and species richness decline in response to habitat loss and fragmentation. Studying variation in bird song structure across modified landscapes can provide insights into the effects of habitat alterations on coherence of social interactions within populations. Here, we tested whether fragmentation or change of habitat quality within box‐ironbark forest of central Victoria impacted cultural connectivity and song characteristics in fuscous honeyeater, a declining common Australian bird. First, we tested whether geographic distance and/or spatially‐explicit landscape connectivity models can explain patterns of song similarity across fragmented landscapes. We found no evidence that distance or habitat fragmentation impacts the nature and transmission of fuscous honeyeater song, and concluded that acoustic connectivity at the scale of our study is high. Second, we tested whether variation in habitat quality explains variation in song characteristics. In accordance with acoustic adaptation to habitat structure, birds sang longer songs in sites with more large trees and produced longer common song elements in sites with greater tree height. However, the acoustic adaptation hypothesis cannot explain the finding that in less‐disturbed landscapes with higher tree‐cover birds sang songs (and song elements) with higher maximum frequency and wider frequency bandwidth. We also found that birds sing longer and more variable songs of wider frequency bandwidth in less disturbed sites with a greater number of large mature trees, which may represent better feeding resources. Our study suggests that changes in song structure with habitat degradation could signal disturbed population processess, such as changes in the acoustic communication among resident birds.     

Introductory gestures before songbird vocal displays are shaped by learning and biological predispositions

Numerous animal displays begin with introductory gestures. For example, lizards start their head-bobbing displays with introductory push-ups, and many songbirds begin their vocal displays by repeating introductory notes (INs) before producing their learned song. Among songbirds, the acoustic structure and the number of INs produced before song vary considerably between individuals in a species. While similar variation in songs between individuals is a result of learning, whether variations in INs are also due to learning remains poorly understood. Here, using natural and experimental tutoring with male zebra finches, we show that mean IN number and IN acoustic structure are learned from a tutor. Interestingly, IN properties and how well INs were learned, were not correlated with the accuracy of song imitation and only weakly correlated with some features of songs that followed. Finally, birds artificially tutored with songs lacking INs still repeated vocalizations that resembled INs, before their songs, suggesting biological predispositions in IN production. These results demonstrate that INs, just like song elements, are shaped both by learning and biological predispositions. More generally, our results suggest mechanisms for generating variation in introductory gestures between individuals while still maintaining the species-specific structure of complex displays like birdsong.     

Randomization Analyses: Mimicry, Geographic Variation and Cultural Evolution of Song in Brood-Parasitic Straw-Tailed Whydahs, Vidua fischeri

Although bird song has been an important model for investigating questions of behavior development, cultural evolution and population differentiation, the quantitative methods of analysis have been problematic. Here we develop and apply quantitative randomization methods to test hypotheses about these processes in a natural population of birds. Songs of the African brood-parasitic straw-tailed whydahs ( Vidua fischeri ) and songs of their host species, the purple grenadier ( Granatina ianthinogaster ), were compared in audiospectrograms for similarity to test the following hypotheses: Whydahs mimic the songs of their host species, they have local song dialects, neighboring males match their song themes, local males match the songs of local hosts, remote populations have different songs according to their geographic distance, and songs undergo cultural evolution over time across generations. Randomization analyses were completed using (1) Mantel matrix statistics and (2) tree-based measures employing Sankoff optimization of Manhattan matrices and approximate randomizations. Our results provide evidence for song mimicry, local song dialects, matching song themes between neighboring males, song matching of local whydah mimics and grenadier song models, correspondence of song differences and geographic distance, and cultural continuity with change in song traditions within a local population. These randomization methods may be useful in other studies of animal communication, and they are sufficiently general for use both with distance matrices derived either from naturalistic impressions of song similarity as in our example or from acoustic measurements.     

The effect of isolation,fragmentation, and population bottlenecks on song structure of a Hawaiian honeycreeper

Little is known about how important social behaviors such as song vary within and among populations for any of the endemic Hawaiian honeycreepers. Habitat loss and non‐native diseases (e.g., avian malaria) have resulted in isolation and fragmentation of Hawaiian honeycreepers within primarily high elevation forests. In this study, we examined how isolation of Hawai'i ‘amakihi (Chlorodrepanis virens) populations within a fragmented landscape influences acoustic variability in song. In the last decade, small, isolated populations of disease tolerant ‘amakihi have been found within low elevation forests, allowing us to record ‘amakihi songs across a large elevational gradient (10–1800 m) that parallels disease susceptibility on Hawai'i island. To understand underlying differences among populations, we examined the role of geographic distance, elevation, and habitat structure on acoustic characteristics of ‘amakihi songs. We found that the acoustic characteristics of ‘amakihi songs and song‐type repertoires varied most strongly across an elevational gradient. Differences in ‘amakihi song types were primarily driven by less complex songs (e.g., fewer frequency changes, shorter songs) of individuals recorded at low elevation sites compared to mid and high elevation populations. The reduced complexity of ‘amakihi songs at low elevation sites is most likely shaped by the effects of habitat fragmentation and a disease‐driven population bottleneck associated with avian malaria, and maintained through isolation, localized song learning and sharing, and cultural drift. These results highlight how a non‐native disease through its influence on population demographics may have also indirectly played a role in shaping the acoustic characteristics of a species.     

Acoustic Divergence with Gene Flow in a Lekking Hummingbird with Complex Songs

Hummingbirds have developed a remarkable diversity of learned vocalizations, from single-note songs to phonologically and syntactically complex songs. In this study we evaluated if geographic song variation of wedge-tailed sabrewings (Campylopterus curvipennis) is correlated with genetic divergence, and examined processes that explain best the origin of intraspecific song variation. We contrasted estimates of genetic differentiation, genetic structure, and gene flow across leks from microsatellite loci of wedge-tailed sabrewings with measures for acoustic signals involved in mating derived from recordings of males singing at leks throughout eastern Mexico. We found a strong acoustic structure across leks and geography, where lek members had an exclusive assemblage of syllable types, differed in spectral and temporal measurements of song, and song sharing decreased with geographic distance. However, neutral genetic and song divergence were not correlated, and measures of genetic differentiation and migration estimates indicated gene flow across leks. The persistence of acoustic structuring in wedge-tailed sabrewings may thus best be explained by stochastic processes across leks, in which intraspecific vocal variation is maintained in the absence of genetic differentiation by postdispersal learning and social conditions, and by geographical isolation due to the accumulation of small differences, producing most dramatic changes between populations further apart.     

MHC‐mediated sexual selection on birdsong: Generic polymorphism,particular alleles and acoustic signals

Several hypotheses predict that the major histocompatibility complex (MHC) drives mating preference in females. Olfactory, colour or morphological traits are often found as reliable signals of the MHC profile, but the role of avian song mediating MHC‐based female choice remains largely unexplored. We investigated the relationship between several MHC and acoustic features in the collared flycatcher (Ficedula albicollis), a European passerine with complex songs. We screened a fragment of the class IIB second exon of the MHC molecule, of which individuals harbour 4–15 alleles, while considerable sequence diversity is maintained at the population level. To make statistical inferences from a large number of comparisons, we adopted both null‐hypothesis testing and effect size framework in combination with randomization procedures. After controlling for potential confounding factors, neither MHC allelic diversity nor the presence of particular alleles was associated remarkably with the investigated qualitative and quantitative song traits. Furthermore, genetic similarity among males based on MHC sequences was not reflected by the similarity in their song based on syllable content. Overall, these results suggest that the relationship between features of song and the allelic composition and diversity of MHC is not strong in the studied species. However, a biologically motivated analysis revealed that individuals that harbour an MHC allele that impairs survival perform songs with broader frequency range. This finding suggests that certain aspects of the song may bear reliable information concerning the MHC profile of the individuals, which can be used by females to optimize mate choice.     

Classification of group-specific variations in songs within House Wren species using machine learning models

Songbirds have shown variation in vocalizations across different populations and different geographical ranges. Such variations can over time lead to divergence in song characteristics, sometimes referred to as dialects. House Wren (Troglodytes aedon) is one such widely distributed bird species that has shown variation in its song characteristics within different populations. Traditionally, such studies have been conducted using manual approaches for classification. In this work we explore the use of machine learning models that can assist in performing classification of bird songs at a conspecific level. Two machine learning techniques, the random forest and a shallow feed forward neural network, are fed with pre-computed sound features to classify vocal variation in House Wren species across different reported population groups and latitudinal areas. A randomized approach is employed to create balanced subsets of sounds from different locations for repeated classification runs in order to provide a reliable estimate of performance. It is observed that such an automated approach is able to classify variations in songs within House Wren with high accuracy. We were also able to confirm the latitudinal variation of House Wren songs reported in previous studies. Given these results, we believe, such a purely data-driven way of analyzing bird songs in general can provide useful hints to biologists on where to look for interesting patterns in order to understand the evolutionary divergence in song characteristics.     

Soft Song in Song Sparrows: Acoustic Structure and Implications for Signal Function

In many species of songbirds, males sometimes produce songs at distinctly lower amplitude than in normal singing. Depending on the species, these 'soft songs' may be sung in the context of female courtship, male–male aggression, or both. In song sparrows, males produce soft songs during aggressive interactions with other males, and the amount of soft song produced is the only singing behavior that can be used to reliably predict a subsequent attack by the singer. Although soft song is clearly an important signal in this species, little is known about the acoustic structure of soft song or about how that structure compares to the structure of normal 'broadcast song'. We recorded a large sample of soft songs and broadcast songs from 10 male song sparrows, and measured song amplitudes in the field while controlling the subject's distance to a calibrated microphone. We show that song sparrow males produce songs over a wide range of amplitudes, with soft songs in the range of 55–77 dB sound pressure level and broadcast songs in the range of 78–85 dB. We present evidence for two types of soft song: 'crystallized' soft songs that are broadcast repertoire song types sung at low amplitude, and 'warbled' soft songs that are not found in the broadcast repertoire. Although highly variable, warbled soft songs produced by individual birds could be grouped into song types based on spectrographic similarity. To our knowledge, a distinct repertoire of soft song types has not been previously reported for any songbird.     

Acoustic characteristics, early experience, and endocrine status interact to modulate female zebra finches' behavioral responses to songs

Female songbirds use male songs as an important criterion for mate selection. Properties of male songs are thought to indicate the male's quality as a potential mate. Song preferences in female zebra finches are known to be influenced by two factors--early auditory experience and the acoustic characteristics of males' songs. Studies often investigate song preferences by priming females with estrogen. However, estrogenic influences on song preferences have not been studied. We investigated the relative influence of early auditory experience, acoustic features of songs, and estrogen availability on song responsiveness in female zebra finches. Juvenile female zebra finches were tutored for 10 days with 40 songs per day with one of three acoustically different song types--simple songs, long-bout songs or complex songs. A fourth group of females was untutored. Aside from this brief song exposure, females were raised and maintained without exposure to male songs. During adulthood, females' behavioral responses to the three song types were tested under three hormone conditions--untreated, estradiol-treated and 1,4,6-androstatriene-3,17-dione (ATD)-treated (to lower endogenous estrogen). Based on the results of our study, four conclusions can be drawn. First, song responsiveness in female zebra finches is strongly affected by minimal early acoustic experience. Second, inexperienced female zebra finches are inherently biased to respond more to complex songs over other song types Third, although female zebra finches are inherently biased to respond more to complex songs, early acoustic experience may either reinforce or weaken this inherent responsiveness to complex songs. Fourth, estrogen selectively accentuates song responsiveness in acoustically-experienced female zebra finches.     

Phylogenetic signal in the song of crests and kinglets (Aves: Regulus)

Territorial song structures are often the most prominent characters for distinguishing closely related taxa among songbirds. Learning processes may cause convergent evolution of passerine songs, but phylogenetic information of acoustic traits can be investigated with the help of molecular phylogenies, which are not affected by cultural evolutionary processes. We used a phylogeny based on cytochrome b sequences to trace the evolution of territorial song within the genus Regulus. Five discrete song units are defined as basic components of regulid song via sonagraphic measurements. Traits of each unit are traced on a molecular tree and a mean acoustic character difference between taxon pairs is calculated. Acoustic divergence between regulid taxa correlates strongly with genetic distances. Syntax features of complete songs and of single units are most consistent with the molecular data, whereas the abundance of certain element types is not. Whether song characters are innate or learned was interpreted using hand-reared birds in aviary experiments. We found that convergent character evolution seems to be most probable for learned acoustic traits. We conclude that syntax traits of whole verses or subunits of territorial song, especially innate song structures, are the most reliable acoustic traits for phylogenetic reconstructions in Regulus.     

Temporal precision and temporal drift in brain and behavior of zebra finch song.

In the zebra finch forebrain nucleus robustus archistriatalis (RA), neurons burst during singing. We showed that the internal structure of spike bursts was regulated with a precision of circa 0.2 ms, and yielded alignment of acoustic features of song with a precision of circa 1 ms. In addition, interburst intervals and corresponding syllable durations displayed systematic variation within song (average elongation 0.3 ms/s song), and slower "drift" across songs. Systematic variation on even a coarser time scale might be difficult to detect in other systems, but could affect the analysis of temporal patterning. The close relationship between precise timing of individual spikes and stereotypic behavior suggests that song is represented in RA by a temporal code.     

Is the Soft Song of the Brownish‐Flanked Bush Warbler an Aggressive Signal?

Soft songs have been detected in many songbirds, but in most species, research on soft songs has lagged behind studies of broadcast songs. In this study, we describe the acoustic features of a soft song in the brownish‐flanked bush warbler Cettia fortipes. Compared with the broadcast song, the warbler's soft song was characterized by a lower minimum frequency and longer duration, and it had a higher proportion of rapid frequency modulation notes. Using playback experiments, some in combination with mounted specimens, we found different responses to soft and broadcast songs, and we found that soft song can predict aggressive escalation (attack). We conclude that the soft song is an aggressive signal in this species.