Automatic sound classification is a promising approach to analyze the spatio-temporal dynamics of natural sound sources. Convolutional neural networks (CNNs) have proved to be particularly effective for the classification of animal sounds. However, even if these algorithms are more efficient than their predecessors, CNNs need to show stable performances despite various sources of variability. The signal-to-noise ratio (SNR) has been identified in the literature as a key parameter in the performance of sound classification algorithms. In this study, we investigated the influence of the SNR during the training and inference of a custom-built CNN for the mono-specific classification of Boreal Owl (Aegolius funereus) vocalizations. The repetitive, stereotyped nocturnal vocalizations of this species are ideal for isolating the effects of the SNR. Experiments showed that even if the custom-built model was trained with very low SNRs, it still very poorly detected low SNR vocalizations during testing. Performance drops per SNR of the custom-built model were comparable to those of the popular model BirdNET. If the custom-built model outperformed BirdNET overall, both models seemed to have similar performance drops as soon as vocalizations had a SNR inferior to 3 dB. Finally, the custom-built model classified 6 years of audio recordings from 4 autonomous recorders in the Risoux forest (France) and highlighted the variations in the Boreal Owl vocalization activity on a yearly, monthly and hourly scale.

Despite hosting some of the highest concentrations of biodiversity and providing invaluable goods and services in the oceans, coral reefs are under threat from global change and other local human impacts. Changes in living ecosystems often induce changes in their acoustic characteristics, but despite recent efforts in passive acoustic monitoring of coral reefs, rapid measurement and identification of changes in their soundscapes remains a challenge. Here we present the new open-source software CoralSoundExplorer, which is designed to study and monitor coral reef soundscapes. CoralSoundExplorer uses machine learning approaches and is designed to eliminate the need to extract conventional acoustic indices. To demonstrate CoralSoundExplorer's functionalities, we use and analyze a set of recordings from three coral reef sites, each with different purposes (undisturbed site, tourist site and boat site), located on the island of Bora-Bora in French Polynesia. We explain the CoralSoundExplorer analysis workflow, from raw sounds to ecological results, detailing and justifying each processing step. We detail the software settings, the graphical representations used for visual exploration of soundscapes and their temporal dynamics, along with the analysis methods and metrics proposed. We demonstrate that CoralSoundExplorer is a powerful tool for identifying disturbances affecting coral reef soundscapes, combining visualizations of the spatio-temporal distribution of sound recordings with new quantification methods to characterize soundscapes at different temporal scales.

Passive acoustic monitoring of biodiversity is growing fast, as it offers an alternative to traditional aural point count surveys, with the possibility to deploy long-term acoustic surveys in large and complex natural environments. However, there is still a clear need to evaluate how the frequency- and distance-dependent attenuation of sound as well as the ambient sound level impact the acoustic detection distance of the soniferous species in natural environments over the diel cycles and across seasons. This is of great importance to avoid pseudoreplication and to provide relevant biodiversity indicators, including species richness, species abundance and species density. To address the issue of detection distance, we tested a field-based protocol in a Neotropical rainforest (French Guiana, France) and in an Alpine coniferous forest (Jura, France). This standardized and repeatable method consists in a recording session of the ambient sound directly followed by an experiment using a calibrated white noise sound broadcast at different positions along a 100 m linear transect. We then used acoustic laws to reveal the basic physics behind sound propagation attenuation. We demonstrate that habitat attenuation in two different kinds of forests can be modelled by an exponential decay law with a linear dependence on frequency and distance. We also report that habitat attenuation, as first approximation, can be summarized by a single value, the coefficient of attenuation of the habitat. Finally, we show that the detection distance can be predicted knowing the contribution of each attenuation factor, the coefficient of attenuation of the habitat, the ambient sound pressure level and the amplitude and frequency bandwidth characteristics of the transmitted sound. We show that the detection distance mostly depends on the ambient sound and may vary by a factor of up to 5 over the diel cycle and across seasons. These results reinforce the need to take into account the variation of the detection distance when performing passive acoustic surveys and producing reliable biodiversity indicators.

One mainstay of soundscape ecology is to understand acoustic pattern changes, in particular the relative balance between biophony (biotic sounds), geophony (abiotic sounds), and anthropophony (human-related sounds). However, little research has been pursued to automatically track these three components. Here, we introduce a 15-year program that aims at estimating soundscape dynamics in relation to possible land use and climate change. We address the relative prevalence patterns of these components during the first year of recording. Using four recorders, we monitored the soundscape of a large coniferous Alpine forest at the France-Switzerland border. We trained an artificial neural network (ANN) with mel frequency cepstral coefficients to systematically detect the occurrence of silence and sounds coming from birds, mammals, insects (biophony), rain (geophony), wind (geophony), and aircraft (anthropophony). The ANN satisfyingly classified each sound type. The soundscape was dominated by anthropophony (75% of all files), followed by geophony (57%), biophony (43%), and silence (14%). The classification revealed expected phenologies for biophony and geophony and a co-occurrence of biophony and anthropophony. Silence was rare and mostly limited to night time. It was possible to track the main soundscape components in order to empirically estimate their relative prevalence across seasons. This analysis reveals that anthropogenic noise is a major component of the soundscape of protected habitats, which can dramatically impact local animal behavior and ecology.

The active space is a central bioacoustic concept to understand communication networks and animal behavior. Propagation of biological acoustic signals has often been studied in homogeneous environments using an idealized circular active space representation, but few studies have assessed the variations of the active space due to environment heterogeneities and transmitter position. To study these variations for mountain birds like the rock ptarmigan, we developed a sound propagation model based on the parabolic equation method that accounts for the topography, the ground effects, and the meteorological conditions. The comparison of numerical simulations with measurements performed during an experimental campaign in the French Alps confirms the capacity of the model to accurately predict sound levels. We then use this model to show how mountain conditions affect surface and shape of active spaces, with topography being the most significant factor. Our data reveal that singing during display flights is a good strategy to adopt for a transmitter to expand its active space in such an environment. Overall, our study brings new perspectives to investigate the spatiotemporal dynamics of communication networks.

The collection and interpretation of field data is a prerequisite for informed conservation in protected environments. Although several techniques, including camera trapping and passive acoustic monitoring, have been developed to estimate the presence of animal species, very few attempts have been made to monitor ecological functions. Pollination by insects and wood use, including tree related foraging and intraspecific communication, by woodpeckers are key functions that need to be assessed in order to better understand and preserve forest ecosystems within the context of climate change. Here, we developed and applied for the first time an acoustic survey to monitor pollination by insects and wood use by woodpeckers in a protected Alpine forest in France. We deployed four autonomous recorders over a year, resulting in 2285 h of recordings. We trained a convolutional neural network (CNN) on spectrographic images to automatically detect the sounds of flying insects' buzzing and woodpeckers' drumming as they forage and call. We used the output of the CNN to estimate the seasonality, diel pattern, climatic breadth and distribution of both functions and their relationships with weather parameters. Our method showed that insects were flying (therefore potentially pollinating flowers) in bright, warm and dry conditions, after dawn and before dusk during spring and summer. Woodpeckers were mainly drumming around March at the time of pair formation in cool and wet conditions. Having considered the role of weather parameters, climate change might have contrasting effects on insect buzzing and woodpecker drumming, with an increase in temperature being favorable to pollination by insects but not to wood use by woodpeckers, and a concomitant increase in relative humidity being favorable to wood use but not to pollination. This study reveals that a systemic facet of biodiversity can be tracked using sound, and that acoustics provide valuable information for the environment description.

Acoustic communication networks among birds are widely studied in homogeneous environment like tropical forest or open field. There is no study in heterogeneous environment to our knowledge. However, the propagation of an acoustic signal is strongly influenced by topography, meteorological and ground effects, especially at long distance. For bird species living in high mountains, these effects can have a prominent impact on how their vocalization spreads. Here we try to redefine the active space notion, using the rock ptarmigan as a model. We develop a sound propagation code dedicated to bioacoustic studies based on the parabolic equation method and taking into account the above mentioned effects. Propagation measurements were carried out at a mountain site to provide information on typical conditions encountered and to test the validity of our model in this context. Using this model, we describe how the mountainous conditions affect active spaces of communication compared to typical homogeneous situations. We further test how singing during display flights can be a good strategy for a transmitter to expand its active space in such a context. A fine modeling of the propagation of acoustic signals is likely to provide important cues for understanding communication networks of mountainous species.

Communicating species identity is a key component of many animal signals. However, whether selection for species recognition systematically increases signal diversity during clade radiation remains debated. Here we show that in woodpecker drumming, a rhythmic signal used during mating and territorial defense, the amount of species identity information encoded remained stable during woodpeckers' radiation. Acoustic analyses and evolutionary reconstructions show interchange among six main drumming types despite strong phylogenetic contingencies, suggesting evolutionary tinkering of drumming structure within a constrained acoustic space. Playback experiments and quantification of species discriminability demonstrate sufficient signal differentiation to support species recognition in local communities. Finally, we only find character displacement in the rare cases where sympatric species are also closely related. Overall, our results illustrate how historical contingencies and ecological interactions can promote conservatism in signals during a clade radiation without impairing the effectiveness of information transfer relevant to inter-specific discrimination.

The cost-effectiveness and reduced human effort employed in setting up acoustic monitoring in the field makes bioacoustics an appealing option for wildlife monitoring. This is especially true for secretive vocal species living in remote places. However, acoustic monitoring still raises questions regarding its reliability when compared to other, human-driven methods. In this study we compare different approaches to count rock ptarmigan males, an alpine bird species which lives at high altitudes. The monitoring of rock ptarmigan populations is traditionally conducted using a point-count protocol, with human observers counting singing males from a set of different points. We assessed the (1) feasibility and (2) reliability of an alternative counting method based on acoustic recordings followed by signal analysis and a dedicated statistical approach to estimate the abundance of males. We then (3) compared the results obtained with this bioacoustics monitoring method with those obtained through the point-count protocol approach over three consecutive years. Acoustic analysis demonstrated that rock ptarmigan vocalizations exhibit an individual stereotypy that can be used to estimate the abundance of males. Simulations, using subsets of our recording dataset, demonstrated that the clustering methods used to discriminate between males based on their vocalizations are sensitive to both the number of recorded signals, as well as the number of individuals to be discriminated. Despite these limitations, we highlight the reliability of the bioacoustics approach, showing that it avoids both observer bias and double counting, contrary to the pointcount protocol where this may occur and impair the data reliability. Overall, our study suggests that bioacoustics monitoring should be used in addition to traditional counting methods to obtain a more accurate estimate of rock ptarmigan abundance within Alpine environments.

The source-filter theory of vocal production supports the idea that acoustic signatures are preferentially coded by the fundamental frequency (source-induced variability) and the distribution of energy among the frequency spectrum (filter-induced variability). By investigating the acoustic parameters supporting individuality in lamb bleats, a vocalization which mediates recognition by ewes, here we show that amplitude modulation - an acoustic feature largely independent of the shape of the acoustic tract - can also be an important cue defining an individual vocal signature. Female sheep (Ovis aries) show an acoustic preference for their own lamb. Although playback experiments have shown that this preference is established soon after birth and relies on a unique vocal signature contained in the bleats of the lamb, the physical parameters that encode this individual identity remained poorly identified. We recorded 152 bleats from 13 fifteen-day-old lambs and analyzed their acoustic structure with four complementary statistical methods (ANOVA, potential for individual identity coding PIC, entropy calculation 2(Hs), discriminant function analysis DFA). Although there were slight differences in the acoustic parameters identified by the four methods, it remains that the individual signature relies on both the temporal and frequency domains. The coding of the identity is thus multi-parametric and integrates modulation of amplitude and energy parameters. Specifically, the contribution of the amplitude modulation is important, together with the fundamental frequency F-0 and the distribution of energy in the frequency spectrum.

To increase the range of modal speech in natural ambient noise, individuals increase their vocal effort and may pass into the 'shouted speech' register. To date, most studies concerning the influence of distance on spoken communication in outdoor natural environments have focused on the 'productive side' of the human ability to tacitly adjust vocal output to compensate for acoustic losses due to sound propagation. Our study takes a slightly different path as it is based on an adaptive speech production/perception experiment. The setting was an outdoor natural soundscape (a plane forest in altitude). The stimuli were produced live during the interaction: each speaker adapted speech to transmit French disyllabic words in isolation to an interlocutor/listener who was situated at variable distances in the course of the experiment (30m, 60m, 90m). Speech recognition was explored by evaluating the ability of 16 normal-hearing French listeners to recognize these words and their constituent vowels and consonants. Results showed that in such conditions, speech adaptation was rather efficient as word recognition remained around 95% at 30m, 85% at 60m and 75% at 90m. We also observed striking differences in patterns of answers along several lines: different distances, speech registers, vowels and consonants.

The song of songbirds is a testosterone-sensitive behavior that is controlled by brain regions expressing androgen receptors. At higher latitudes, seasonal singing is stimulated by increasing day-length and elevated circulating testosterone. However, a large number of songbird species inhabit equatorial regions under a nearly constant photoperiod, and the neuroendocrine mechanisms of seasonal song in these species have rarely been investigated. We studied males from an equatorial population of the silver-beaked tanager (Ramphocelus carbo), an Amazonian songbird. We found seasonality in dawn-song behavior, which was displayed continuously for more than half a year throughout an extended breeding territoriality stage. The seasonal activation of dawn-song was correlated with an increased area of androgen receptor expression in HVC, a major brain area of song control. However, testosterone levels remained low for several weeks after activation of dawn-song. Circulating levels of testosterone were elevated only later in the breeding season, coinciding with a higher dawn-song output and with the mating period. Our results suggest that the seasonal activation of dawn-song and territoriality involves an increase of androgen target cells in HVC. This mechanism could potentially function to circumvent adverse effects of high testosterone levels in a species with an extended breeding season.

Needs for objective and automatic assessment of animal welfare are increasing in livestock production. The strong dam-offspring relationship is a good model to investigate the vocal expression of emotions in animals since removing the lamb induces behavioural distress in ewes (e.g. agitation and vocalisations). We analysed acoustic characteristics in bleats of ewes following an unpredictable separation from their lambs, in addition to their behavioural and physiological responses. Twenty four ewes of INRA 401 breed were individually exposed to 3 successive 3 min phases: 1) in contact with their lamb placed behind a grid, 2) at 6m away from the lamb and 3) again in contact. Behaviour, vocalisations and cardiac activity (via adhesive external electrodes and a telemetric remote system) were recorded and blood was collected just after the test by venipuncture for cortisol assay. After the lamb was moved away, the ewes were more active and more vigilant, they also bleated more, and their heart rate and cortisol levels dramatically increased. These results confirm that the separation induced behavioural and physiological responses of distress in ewes which reflect negative emotional states. Beside, acoustic analyses show that lamb withdrawal induced changes in the voice characteristics of mothers in correlation with their physiological and behavioural responses. Temporal parameters (e.g. total duration) increased, as did amplitude parameters (e.g. energy, RMS). Concerning the frequency parameters, fundamental frequency increased but frequency Bandwidth, Quartiles (Q25%), (Q50%), (Q75%) decreased. Therefore some bleat characteristics can be used as acoustic markers of emotional distress. Acoustic sensory modality could thus provide an objective basis of negative emotions in sheep, which measure could be easier to automate than other behavioural and physiological correlates of emotional reactions.

The present study aimed to evaluate an experimental approach to individually assess social reactivity among sheep. INRA401 male lambs (n=163) were reared together outdoors as part of a larger flock. Fifteen days after weaning the animals were individually exposed to an arena test of 2 phases (1-social attraction, 2-social isolation) during which proximity toward conspecifics and vocal and locomotor reactivity were measured. One day after the test their inter-individual distances were measured when grazing over a 2-h period in order to estimate their sociability on pasture. This was made using scan sampling recording the identity of the nearest neighbour for each individual, which led to the establishment of a sociability index. Overall, we found that high-pitched bleats recorded during the attraction phase (r=0.22) and the isolation phase (r=0.23) of the arena test as well as the locomotor activity measured during the isolation phase (r=0.27) were positively correlated with the sociability index. Furthermore, the behaviour of lambs during the isolation phase of the arena test (i.e. vocal and locomotor agitation) appeared to be a significant predictor explaining 13% of the variance of the sociability on pasture. The behavioural reactivity measured through the arena test thus reflects at least to some extent the sociability of sheep. Those results are very encouraging as they suggest that the sociability of lambs could indeed be evaluated through a short experimental test, which is less time consuming than field ethological observations.

ABSTRACT In domestic sheep Ovis aries, the mother and the young display a preferential bond for each other that relies on multimodal inter-individual recognition. Lambs show a preference for their own dam shortly after birth, and this is important for their survival. The role of acoustic cues in this early preference for the mother is not clear. The aim of the present work was to analyze the timing of acoustic recognition of the mother and to identify the physical parameters used in the recognition of maternal bleats by the lamb. In a first study, we investigated the ability of lambs to discriminate between the bleats of their own mother and an alien equivalent mother in a two-choice test. Both ewes were hidden behind a canvas sheet and lambs were not allowed to approach the dams closer than 1 m, thus preventing visual as well as olfactory perception. Tests were conducted 12 hr, 24 hr or 48 hr after birth (n=19 or 20/group). An indication of vocal discrimination was already present at 24 hr and at 48 hr lambs spent significantly more time near their mother than near the alien dam. In a second step, we investigated which physical parameters of the bleats were important for recognition. For this, we conducted playback experiments with modified bleats at two weeks postpartum. Ours results show that lambs pay attention to a combination of various time, energy and frequency parameters: timbre (distribution of energy within the spectrum), amplitude and frequency modulations appear to be the most important parameters encoding the individual signature. We conclude that vocal recognition between the ewe and her lamb plays an important role in the display of preferential mother-young bond from very early on. Our studies also demonstrate that the encoding of the individual signature is not limited to the frequency domain but rather involves a multiparametric encoding process.