The machine learning-powered BirdNET App reduces barriers to global bird research by enabling citizen science participation.

The BirdNET App, a free bird sound identification app for Android and iOS that includes over 3,000 bird species, reduces barriers to citizen science while generating tens of millions of bird observations globally that can be used to replicate known patterns in avian ecology.

Characterizing soundscapes across diverse ecosystems using a universal acoustic feature set.

Natural habitats are being impacted by human pressures at an alarming rate. Monitoring these ecosystem-level changes often requires labor-intensive surveys that are unable to detect rapid or unanticipated environmental changes. Here we have developed a generalizable, data-driven solution to this challenge using eco-acoustic data. We exploited a convolutional neural network to embed soundscapes from a variety of ecosystems into a common acoustic space. In both supervised and unsupervised modes, this allowed us to accurately quantify variation in habitat quality across space and in biodiversity through time. On the scale of seconds, we learned a typical soundscape model that allowed automatic identification of anomalous sounds in playback experiments, providing a potential route for real-time automated detection of irregular environmental behavior including illegal logging and hunting. Our highly generalizable approach, and the common set of features, will enable scientists to unlock previously hidden insights from acoustic data and offers promise as a backbone technology for global collaborative autonomous ecosystem monitoring efforts.

Using deep learning to track time × frequency whistle contours of toothed whales without human-annotated training data.

Many odontocetes produce whistles that feature characteristic contour shapes in spectrogram representations of their calls. Automatically extracting the time × frequency tracks of whistle contours has numerous subsequent applications, including species classification, identification, and density estimation. Deep-learning-based methods, which train models using analyst-annotated whistles, offer a promising way to reliably extract whistle contours. However, the application of such methods can be limited by the significant amount of time and labor required for analyst annotation. To overcome this challenge, a technique that learns from automatically generated pseudo-labels has been developed. These annotations are less accurate than those generated by human analysts but more cost-effective to generate. It is shown that standard training methods do not learn effective models from these pseudo-labels. An improved loss function designed to compensate for pseudo-label error that significantly increases whistle extraction performance is introduced. The experiments show that the developed technique performs well when trained with pseudo-labels generated by two different algorithms. Models trained with the generated pseudo-labels can extract whistles with an F1-score (the harmonic mean of precision and recall) of 86.31% and 87.2% for the two sets of pseudo-labels that are considered. This performance is competitive with a model trained with 12 539 expert-annotated whistles (F1-score of 87.47%).

Source separation with an acoustic vector sensor for terrestrial bioacoustics.

Passive acoustic monitoring is emerging as a low-cost, non-invasive methodology for automated species-level population surveys. However, systems for automating the detection and classification of vocalizations in complex soundscapes are significantly hindered by the overlap of calls and environmental noise. We propose addressing this challenge by utilizing an acoustic vector sensor to separate contributions from different sound sources. More specifically, we describe and implement an analytical pipeline consisting of (1) calculating direction-of-arrival, (2) decomposing the azimuth estimates into angular distributions for individual sources, and (3) numerically reconstructing source signals. Using both simulation and experimental recordings, we evaluate the accuracy of direction-of-arrival estimation through the active intensity method (AIM) against the baselines of white noise gain constraint beamforming (WNC) and multiple signal classification (MUSIC). Additionally, we demonstrate and compare source signal reconstruction with simple angular thresholding and a wrapped Gaussian mixture model. Overall, we show that AIM achieves higher performance than WNC and MUSIC, with a mean angular error of about 5°, robustness to environmental noise, flexible representation of multiple sources, and high fidelity in source signal reconstructions.

Accounting for the Lombard effect in estimating the probability of detection in passive acoustic surveys: Applications for single sensor mitigation and monitoring.

The detection range of calling animals is commonly described by the passive sonar equations. However, the sonar equations do not account for interactions between source and ambient sound level, i.e., the Lombard effect. This behavior has the potential to introduce non-linearities into the sonar equations and result in incorrectly predicted detection ranges. Here, we investigate the relationship between ambient sound and effective detection ranges for North Atlantic right whales (Eubalaena glacialis) in Cape Cod Bay, MA, USA using a sparse array of acoustic recorders. Generalized estimating equations were used to model the probability that a call was detected as a function of distance between the calling animal and the sensor and the ambient sound level. The model suggests a non-linear relationship between ambient sound levels and the probability of detecting a call. Comparing the non-linear model to the linearized version of the same model resulted in 12 to 25% increases in the effective detection range. We also found evidence of the Lombard effect suggesting that it is the most plausible cause for the non-linearity in the relationship. Finally, we suggest a simple modification to the sonar equation for estimating detection probability for single sensor monitoring applications.

Detection probability and density estimation of fin whales by a Seaglider.

A single-hydrophone ocean glider was deployed within a cabled hydrophone array to demonstrate a framework for estimating population density of fin whales (Balaenoptera physalus) from a passive acoustic glider. The array was used to estimate tracks of acoustically active whales. These tracks became detection trials to model the detection function for glider-recorded 360-s windows containing fin whale 20-Hz pulses using a generalized additive model. Detection probability was dependent on both horizontal distance and low-frequency glider flow noise. At the median 40-Hz spectral level of 97 dB re 1 µPa2/Hz, detection probability was near one at horizontal distance zero with an effective detection radius of 17.1 km [coefficient of variation (CV) = 0.13]. Using estimates of acoustic availability and acoustically active group size from tagged and tracked fin whales, respectively, density of fin whales was estimated as 1.8 whales per 1000 km2 (CV = 0.55). A plot sampling density estimate for the same area and time, estimated from array data alone, was 1.3 whales per 1000 km2 (CV = 0.51). While the presented density estimates are from a small demonstration experiment and should be used with caution, the framework presented here advances our understanding of the potential use of gliders for cetacean density estimation.

Silbido profundo: An open source package for the use of deep learning to detect odontocete whistles.

This work presents an open-source matlab software package for exploiting recent advances in extracting tonal signals from large acoustic data sets. A whistle extraction algorithm published by Li, Liu, Palmer, Fleishman, Gillespie, Nosal, Shiu, Klinck, Cholewiak, Helble, and Roch [(2020). Proceedings of the International Joint Conference on Neural Networks, July 19-24, Glasgow, Scotland, p. 10] is incorporated into silbido, an established software package for extraction of cetacean tonal calls. The precision and recall of the new system were over 96% and nearly 80%, respectively, when applied to a whistle extraction task on a challenging two-species subset of a conference-benchmark data set. A second data set was examined to assess whether the algorithm generalized to data that were collected across different recording devices and locations. These data included 487 h of weakly labeled, towed array data collected in the Pacific Ocean on two National Oceanographic and Atmospheric Administration (NOAA) cruises. Labels for these data consisted of regions of toothed whale presence for at least 15 species that were based on visual and acoustic observations and not limited to whistles. Although the lack of per whistle-level annotations prevented measurement of precision and recall, there was strong concurrence of automatic detections and the NOAA annotations, suggesting that the algorithm generalizes well to new data.

Limited vocal compensation for elevated ambient noise in bearded seals: implications for an industrializing Arctic Ocean.

Vocalizing animals have several strategies to compensate for elevated ambient noise. These behaviours evolved under historical conditions, but compensation limits are quickly being reached in the Anthropocene. Acoustic communication is essential to male bearded seals that vocalize for courtship and defending territories. As Arctic sea ice declines, industrial activities and associated anthropogenic noise are likely to increase. Documenting how seals respond to noise and identifying naturally occurring behavioural thresholds would indicate either their resilience or vulnerability to changing soundscapes. We investigated whether male bearded seals modified call amplitudes in response to changing ambient noise levels. Vocalizing seals increased their call amplitudes until ambient noise levels reached an observable threshold, above which call source levels stopped increasing. The presence of a threshold indicates limited noise compensation for seals, which still renders them vulnerable to acoustic masking of vocal signals. This behavioural threshold and response to noise is critical for developing management plans for an industrializing Arctic.

Using the ecological significance of animal vocalizations to improve inference in acoustic monitoring programs.

Recent bioacoustic advances have facilitated large-scale population monitoring for acoustically active species. Animal sounds, however, can of information that is underutilized in typical approaches to passive acoustic monitoring (PAM) that treat sounds simply as detections. We developed 3 methods of extracting additional ecological detail from acoustic data that are applicable to a broad range of acoustically active species. We conducted landscape-scale passive acoustic surveys of a declining owl species and an invasive congeneric competitor in California. We then used sex-specific vocalization frequency to inform multistate occupancy models; call rates at occupied sites to characterize interactions with interspecific competitors and assess habitat quality; and a flexible multivariate approach to differentiate individuals based on vocal characteristics. The multistate occupancy models yielded novel estimates of breeding status occupancy rates that were more robust to false detections and captured known habitat associations more consistently than single-state occupancy models agnostic to sex. Call rate was related to the presence of a competitor but not habitat quality and thus could constitute a useful behavioral metric for interactions that are challenging to detect in an occupancy framework. Quantifying multivariate distance between groups of vocalizations provided a novel quantitative means of discriminating individuals with ≥20 vocalizations and a flexible tool for balancing type I and II errors. Therefore, it appears possible to estimate site turnover and demographic rates, rather than just occupancy metrics, in PAM programs. Our methods can be applied individually or in concert and are likely generalizable to many acoustically active species. As such, they are opportunities to improve inferences from PAM data and thus benefit conservation.

Uso de la Importancia Ecológica de las Vocalizaciones Animales para Mejorar la Inferencia en los Programas de Monitoreo Acústico Resumen Los avances bioacústicos recientes han facilitado el monitoreo a gran escala de poblaciones de especies acústicamente activas. Sin embargo, los sonidos de animales pueden transmitir cantidades sustanciales de información que queda utilizada insuficientemente en las estrategias comunes de monitoreo acústico pasivo (MAP) que tratan a los sonidos como simples detecciones. Desarrollamos tres métodos de extracción de detalles ecológicos adicionales de los datos acústicos que son aplicables a una gama amplia de especies acústicamente activas. Realizamos censos acústicos pasivos a escala de paisaje para una especie de búho en declinación y para un competidor congenérico invasivo en California. Después utilizamos la frecuencia de vocalizaciones específicas por sexo para orientar los modelos multiestado de ocupación; las tasas de llamados en sitios ocupados para caracterizar las interacciones con los competidores interespecíficos y evaluar la calidad de su hábitat; y una estrategia multivariada flexible para diferenciar a los individuos con base en sus características vocales. Los modelos multiestado de ocupación brindaron estimaciones novedosas para las tasas de ocupación por estado reproductivo que fueron más sólidas ante las detecciones falsas y capturaron el número de asociaciones de hábitat más sistemáticamente que los modelos de estado único agnósticos al sexo. La tasa de llamados estuvo relacionada con la presencia de un competidor pero no con la calidad del hábitat y por lo tanto podría constituir una medida conductual útil para las interacciones que son difíciles de detectar en un marco de trabajo de ocupación. La cuantificación de la distancia multivariada entre los grupos de vocalizaciones proporcionó un medio cuantitativo novedoso para discriminar a los individuos con ≥20 vocalizaciones y una herramienta flexible para balancear los errores del tipo I y del tipo II. Por lo tanto, parecer que hay posibilidad de estimar las tasas demográficas y de rotación, en lugar de sólo las medidas de ocupación, en los programas MAP. Nuestros métodos pueden aplicarse individualmente o de manera conjunta y es probable poder generalizarlas para muchas especies acústicamente activas. Dicho así, son oportunidades para mejorar las inferencias de los datos MAP y por lo tanto, beneficiar a la conservación.

Exploring movement patterns and changing distributions of baleen whales in the western North Atlantic using a decade of passive acoustic data.

Six baleen whale species are found in the temperate western North Atlantic Ocean, with limited information existing on the distribution and movement patterns for most. There is mounting evidence of distributional shifts in many species, including marine mammals, likely because of climate-driven changes in ocean temperature and circulation. Previous acoustic studies examined the occurrence of minke (Balaenoptera acutorostrata) and North Atlantic right whales (NARW; Eubalaena glacialis). This study assesses the acoustic presence of humpback (Megaptera novaeangliae), sei (B. borealis), fin (B. physalus), and blue whales (B. musculus) over a decade, based on daily detections of their vocalizations. Data collected from 2004 to 2014 on 281 bottom-mounted recorders, totaling 35,033 days, were processed using automated detection software and screened for each species' presence. A published study on NARW acoustics revealed significant changes in occurrence patterns between the periods of 2004-2010 and 2011-2014; therefore, these same time periods were examined here. All four species were present from the Southeast United States to Greenland; humpback whales were also present in the Caribbean. All species occurred throughout all regions in the winter, suggesting that baleen whales are widely distributed during these months. Each of the species showed significant changes in acoustic occurrence after 2010. Similar to NARWs, sei whales had higher acoustic occurrence in mid-Atlantic regions after 2010. Fin, blue, and sei whales were more frequently detected in the northern latitudes of the study area after 2010. Despite this general northward shift, all four species were detected less on the Scotian Shelf area after 2010, matching documented shifts in prey availability in this region. A decade of acoustic observations have shown important distributional changes over the range of baleen whales, mirroring known climatic shifts and identifying new habitats that will require further protection from anthropogenic threats like fixed fishing gear, shipping, and noise pollution.

Acoustically advertising male harbour seals in southeast Alaska do not make biologically relevant acoustic adjustments in the presence of vessel noise.

Aquatically breeding harbour seal (Phoca vitulina) males use underwater vocalizations during the breeding season to establish underwater territories, defend territories against intruder males, and possibly to attract females. Vessel noise overlaps in frequency with these vocalizations and could negatively impact breeding success by limiting communication space. In this study, we investigated whether harbour seals employed anti-masking strategies to maintain communication in the presence of vessel noise in Glacier Bay National Park and Preserve, Alaska. Harbour seals in this location did not sufficiently adjust source levels or acoustic parameters of vocalizations to compensate for acoustic masking. Instead, for every 1 dB increase in ambient noise, signal excess decreased by 0.84 dB, indicating a reduction in communication space when vessels passed. We suggest that harbour seals may already be acoustically advertising at or near a biologically maximal sound level and therefore lack the ability to increase call amplitude to adjust to changes in their acoustic environment. This may have significant implications for this aquatically breeding pinniped, particularly for populations in high noise regions.

High-intensity urban light installation dramatically alters nocturnal bird migration.

Billions of nocturnally migrating birds move through increasingly photopolluted skies, relying on cues for navigation and orientation that artificial light at night (ALAN) can impair. However, no studies have quantified avian responses to powerful ground-based light sources in urban areas. We studied effects of ALAN on migrating birds by monitoring the beams of the National September 11 Memorial & Museum's "Tribute in Light" in New York, quantifying behavioral responses with radar and acoustic sensors and modeling disorientation and attraction with simulations. This single light source induced significant behavioral alterations in birds, even in good visibility conditions, in this heavily photopolluted environment, and to altitudes up to 4 km. We estimate that the installation influenced ≈1.1 million birds during our study period of 7 d over 7 y. When the installation was illuminated, birds aggregated in high densities, decreased flight speeds, followed circular flight paths, and vocalized frequently. Simulations revealed a high probability of disorientation and subsequent attraction for nearby birds, and bird densities near the installation exceeded magnitudes 20 times greater than surrounding baseline densities during each year's observations. However, behavioral disruptions disappeared when lights were extinguished, suggesting that selective removal of light during nights with substantial bird migration is a viable strategy for minimizing potentially fatal interactions among ALAN, structures, and birds. Our results also highlight the value of additional studies describing behavioral patterns of nocturnally migrating birds in powerful lights in urban areas as well as conservation implications for such lighting installations.

A performance comparison of tonal detectors for low-frequency vocalizations of Antarctic blue whales.

Extraction of tonal signals embedded in background noise is a crucial step before classification and separation of low-frequency sounds of baleen whales. This work reports results of comparing five tonal detectors, namely the instantaneous frequency estimator, YIN estimator, harmonic product spectrum, cost-function-based detector, and ridge detector. Comparisons, based on a low-frequency adaptation of the Silbido scoring feature, employ five metrics, which quantify the effectiveness of these detectors to retrieve tonal signals that have a wide range of signal to noise ratios (SNRs) and the quality of the detection results. Ground-truth data were generated by embedding 20 synthetic Antarctic blue whale (Balaenoptera musculus intermedia) calls in randomly extracted 30-min noise segments from a 79 h-library recorded by an Ocean Bottom Seismometer in the Indian Ocean during 2012-2013. Monte-Carlo simulations were performed using 20 trials per SNR, ranging from 0 dB to 15 dB. Overall, the tonal detection results show the superiority of the cost-function-based and the ridge detectors, over the other detectors, for all SNR values. More particularly, for lower SNRs (⩽3 dB), these two methods outperformed the other three with high recall, low fragmentation, and high coverage scores. For SNRs ⩾7 dB, the five methods performed similarly.

Multi-year occurrence of sei whale calls in North Atlantic polar waters.

In 2009-2014, autonomous hydrophones were deployed on established long-term moorings in the Fram Strait and Greenland Sea to record multi-year, seasonal occurrence of vocalizing cetaceans. Sei whales have rarely been observed north of ∼72°N, yet there was acoustic evidence of sei whale presence in the Fram Strait for several months during all five years of the study. More sei whale calls were recorded at the easternmost moorings in the Fram Strait, likely because of the presence of warm Atlantic water and a strong front concentrating prey in this area. Sei whale vocalizations were not recorded at the Greenland Sea 2009-2010 mooring, either because this area is not part of the northward migratory path of sei whales or because oceanographic conditions were not suitable for foraging. No clear relationship between whale presence and water temperature data collected coincident with acoustic data was observed, but decadal time series of water temperature data collected in the eastern Fram Strait by others exhibit a warming trend, which may make conditions suitable for sei whales. Continued monitoring of the region will be required to determine if the presence of sei whales in these polar waters is ephemeral or a common occurrence.

Comparison of fin whale 20 Hz call detections by deep-water mobile autonomous and stationary recorders.

Acoustically equipped deep-water mobile autonomous platforms can be used to survey for marine mammals over intermediate spatiotemporal scales. Direct comparisons to fixed recorders are necessary to evaluate these tools as passive acoustic monitoring platforms. One glider and two drifting deep-water floats were simultaneously deployed within a deep-water cabled hydrophone array to quantitatively assess their survey capabilities. The glider was able to follow a pre-defined track while float movement was somewhat unpredictable. Fin whale (Balaenoptera physalus) 20 Hz pulses were recorded by all hydrophones throughout the two-week deployment. Calls were identified using a template detector, which performed similarly across recorder types. The glider data contained up to 78% fewer detections per hour due to increased low-frequency flow noise present during glider descents. The glider performed comparably to the floats and fixed recorders at coarser temporal scales; hourly and daily presence of detections did not vary by recorder type. Flow noise was related to glider speed through water and dive state. Glider speeds through water of 25 cm/s or less are suggested to minimize flow noise and the importance of glider ballasting, detector characterization, and normalization by effort when interpreting glider-collected data and applying it to marine mammal density estimation are discussed.

Diving behavior of Cuvier's beaked whales inferred from three-dimensional acoustic localization and tracking using a nested array of drifting hydrophone recorders.

Echolocation pulses from Cuvier's beaked whales are used to track the whales' three-dimensional diving behavior in the Catalina Basin, California. In 2016, five 2-element vertical hydrophone arrays were suspended from the surface and drifted at ∼100-m depth. Cuvier's beaked whale pulses were identified, and vertical detection angles were estimated from time-differences-of-arrival of either direct-path signals received on two hydrophones or direct-path and surface-reflected signals received on the same hydrophone. A Bayesian state-space model is developed to track the diving behavior. The model is fit to these detection angle estimates from at least four of the drifting vertical arrays. Results show that the beaked whales were producing echolocation pulses and are presumed to be foraging at a mean depth of 967 m (standard deviation = 112 m), approximately 300 m above the bottom in this basin. Some whales spent at least some time at or near the bottom. Average swim speed was 1.2 m s-1, but swim direction varied during a dive. The average net horizontal speed was 0.6 m s-1. Results are similar to those obtained from previous tagging studies of this species. These methods may allow expansion of dive studies to other whale species that are difficult to tag.

Source levels of foraging humpback whale calls.

Humpback whales produce a wide range of low- to mid-frequency vocalizations throughout their migratory range. Non-song "calls" dominate this species' vocal repertoire while on high-latitude foraging grounds. The source levels of 426 humpback whale calls in four vocal classes were estimated using a four-element planar array deployed in Glacier Bay National Park and Preserve, Southeast Alaska. There was no significant difference in source levels between humpback whale vocal classes. The mean call source level was 137 dBRMS re 1 µPa @ 1 m in the bandwidth of the call (range 113-157 dBRMS re 1 µPa @ 1 m), where bandwidth is defined as the frequency range from the lowest to the highest frequency component of the call. These values represent a robust estimate of humpback whale source levels on foraging grounds and should append earlier estimates.

Source levels and call parameters of harbor seal breeding vocalizations near a terrestrial haulout site in Glacier Bay National Park and Preserve.

Source levels of harbor seal breeding vocalizations were estimated using a three-element planar hydrophone array near the Beardslee Islands in Glacier Bay National Park and Preserve, Alaska. The average source level for these calls was 144 dBRMS re 1 µPa at 1 m in the 40-500 Hz frequency band. Source level estimates ranged from 129 to 149 dBRMS re 1 µPa. Four call parameters, including minimum frequency, peak frequency, total duration, and pulse duration, were also measured. These measurements indicated that breeding vocalizations of harbor seals near the Beardslee Islands of Glacier Bay National Park are similar in duration (average total duration: 4.8 s, average pulse duration: 3.0 s) to previously reported values from other populations, but are 170-220 Hz lower in average minimum frequency (78 Hz).

A complex baleen whale call recorded in the Mariana Trench Marine National Monument.

In fall 2014 and spring 2015, passive acoustic data were collected via autonomous gliders east of Guam in an area that included the Mariana Trench Marine National Monument. A short (2-4 s), complex sound was recorded that features a ∼38 Hz moan with both harmonics and amplitude modulation, followed by broad-frequency metallic-sounding sweeps up to 7.5 kHz. This sound was recorded regularly during both fall and spring surveys. Aurally, the sound is quite unusual and most resembles the minke whale "Star Wars" call. It is likely this sound is biological and produced by a baleen whale.