Stronger together: Combining automated classifiers with manual post-validation optimizes the workload vs reliability trade-off of species identification in bat acoustic surveys

Owing to major technological advances, bioacoustics has become a burgeoning field in ecological research worldwide. Autonomous passive acoustic recorders are becoming widely used to monitor aerial insectivorous bats, and automatic classifiers have emerged to aid researchers in the daunting task of analysing the resulting massive acoustic datasets. However, the scarcity of comprehensive reference call libraries still hampers their wider application in highly diverse tropical assemblages. Capitalizing on a unique acoustic dataset of >650,000 bat call sequences collected over a 3-year period in the Brazilian Amazon, the aims of this study were (a) to assess how pre-identified recordings of free-flying and hand-released bats could be used to train an automatic classification algorithm (random forest), and (b) to optimize acoustic analysis protocols by combining automatic classification with visual post-validation, whereby we evaluated the proportion of sound files to be post-validated for different thresholds of classification accuracy. Classifiers were trained at species or sonotype (group of species with similar calls) level. Random forest models confirmed the reliability of using calls of both free-flying and hand-released bats to train custom-built automatic classifiers. To achieve a general classification accuracy of ~85%, random forest had to be trained with at least 500 pulses per species/sonotype. For seven out of 20 sonotypes, the most abundant in our dataset, we obtained high classification accuracy (>90%). Adopting a desired accuracy probability threshold of 95% for the random forest classifier, we found that the percentage of sound files required for manual post-validation could be reduced by up to 75%, a significant saving in terms of workload. Combining automatic classification with manual ID through fully customizable classifiers implemented in open-source software as demonstrated here shows great potential to help overcome the acknowledged risks and biases associated with the sole reliance on automatic classification.     

Acoustic characterization of bats from Malta: setting a baseline for monitoring and conservation of bat populations

Bioacoustic research has made several advancements in developing systems to record extensive acoustic data and classify bat echolocation calls to species level using automated classifiers. These systems are useful as echolocation calls give valuable information on bat behaviour and ecology and hence are widely used for research and conservation of bat populations. Despite the challenges associated with automated classifiers, due to the interspecific differences in call characteristics of bat species found in the Maltese Islands, the use of a quantitative and automated approach is investigated. The sound analysis pipeline involved the use of an algorithm to clean sound files from background noise and measure temporal and spectral parameters of bat echolocation calls. These parameters were then fed to a trained and validated artificial neural network using a bat call library built from reference bat calls from Malta. The automatic classifier achieved an overall correct classification rate of 98%. This high correct classification rate for reliable species identification may have benefitted from the absence of typically problematic species, such as species in the genus Myotis, in the analyses. This study’s results pave the way for efficient and reliable bat acoustic surveys in Malta in aid of necessary monitoring and conservation by providing an updated bat species list and their echolocation characteristics.     

Echolocation call description of 15 species of Middle-Eastern desert dwelling insectivorous bats

Modern advances in acoustic technology have made possible new and broad ranges of research in bioacoustics, particularly with regard to echolocating bats. In the present study, we present an acoustic guide to the calls of 15 species of bats in the Arava rift valley, Israel, with a focus on their bioacoustics, habitat use and explaining differences between similar species. We also describe a potential case of frequency separation where four bat species using six call types appear to separate the frequencies of their calls to minimize overlap. The studied community of bat species is also found in other Middle Eastern deserts including the deserts of Jordan, Syria and Saudi Arabia and we hope that data gathered will benefit other bat researchers in the region.     

Automated classification of bat echolocation call recordings with artificial intelligence

Acoustic recorders are commonly used to remotely monitor and collect data on bats (Order Chiroptera). These efforts result in many acoustic recordings that must be classified by a bat biologist with expertise in call classification in order to obtain useful information. The rarity of this expertise and time constraints have prompted efforts to automatically classify bat species in acoustic recordings using a variety of learning methods. There are several software programs available for this purpose, but they are imperfect and the United States Fish and Wildlife Service often recommends that a qualified acoustic analyst review bat call identifications even if using these software programs. We sought to build a model to classify bat species using modern computer vision techniques. We used images of bat echolocation calls (i.e., plots of the pulses) to train deep learning computer vision models that automatically classify bat calls to species. Our model classifies 10 species, five of which are protected under the Endangered Species Act. We evaluated our models using standard model validation procedures, and performed two external tests. For these tests, an entire dataset was withheld from the procedure before splitting the data into training and validation sets. We found that our validation accuracy (92%) and testing accuracy (90%) were higher than when we used Kaleidoscope Pro and BCID software (65% and 61% accuracy, respectively) to evaluate the same calls. Our results suggest that our approach is effective at classifying bat species from acoustic recordings, and our trained model will be incorporated into new bat call identification software: WEST-EchoVision.     

Acoustic classification of dolphins in the California Current using whistles,echolocation clicks,and burst pulses

Passive acoustic monitoring of dolphins is limited by our ability to classify calls to species. Significant overlap in call characteristics among many species, combined with a wide range of call types and acoustic behavior, makes classification of calls to species challenging. Here, we introduce BANTER, a compound acoustic classification method for dolphins that utilizes information from all call types produced by dolphins rather than a single call type, as has been typical for acoustic classifiers. Output from the passive acoustic monitoring software, PAMGuard, was used to create independent classifiers for whistles, echolocation clicks, and burst pulses, which were then merged into a final, compound classifier for each species. Classifiers for five species found in the California Current ecosystem were trained and tested using 153 single‐species acoustic events recorded during a 4.5 mo combined visual and acoustic shipboard cetacean survey off the west coast of the United States. Correct classification scores for individual species ranged from 71% to 92%, with an overall correct classification score of 84% for all five species. The conceptual framework of this approach easily lends itself to other species and study areas as well as to noncetacean taxa.     

Acoustic identification of bats in the eastern United States: A comparison of parametric and nonparametric methods

Ultrasonic detectors are widely used to survey bats in ecological studies. To evaluate efficacy of acoustic identification, we compiled a library of search phase calls from across the eastern United States using the Anabat system. The call library included 1,846 call sequences of 12 species recorded from 14 states. We determined accuracy rates using 3 parametric and 4 nonparametric classification functions for acoustic identification. The 2 most flexible classification functions also were the most accurate: neural networks (overall classification accuracy = 0.94) and mixture discriminant analysis incorporating an adaptive regression model (overall classification accuracy = 0.93). Flexible nonparametric methods offer substantial benefits when discriminating among closely related species and may preclude the need to group species with similar calls. We demonstrate that quantitative methods provide an effective technique to acoustically identify bats in the eastern United States with known accuracy rates. © 2011 The Wildlife Society.     

The use of automated identification of bat echolocation calls in acoustic monitoring: A cautionary note for a sound analysis

Bats are a species-rich order of mammals providing key ecosystem services. Because bats are threatened by human action and also serve as important bioindicators, monitoring their populations is of utmost importance. However, surveying bats is difficult because of their nocturnal habits, elusiveness and sensitivity to disturbance. Bat detectors allow echolocating bats to be surveyed non-invasively and record species that would otherwise be difficult to observe by capture or roost inspection. Unfortunately, several bat species cannot be identified confidently from their calls so acoustic classification remains ambiguous or impossible in some cases.The popularity of automated classifiers of bat echolocation calls has escalated rapidly, including that of several packages available on purchase. Such products have filled a vacant niche on the market mostly in relation to the expanding monitoring efforts related to the development of wind energy production worldwide.We highlight that no classifier has yet proven capable of providing correct classifications in 100% of cases or getting close enough to this ideal performance. Besides, from the literature available and our own experience we argue that such tools have not yet been tested sufficiently in the field. Visual inspection of calls whose automated classification is judged suspicious is often recommended, but human intervention a posteriori represents a circular argument and requires noticeable experience.We are concerned that neophytes – including consultants with little experience with bats but specialized into other taxonomical groups – will accept passively automated responses of tools still awaiting sufficient validation. We remark that bat call identification is a serious practical issue because biases in the assessment of bat distribution or habitat preferences may lead to wrong management decisions with serious conservation consequences. Automated classifiers may crucially aid bat research and certainly merit further investigations but the boost in commercially available software may have come too early. Thorough field tests need to be carried out to assess limitations and strengths of these tools.     

An automated multispecies bioacoustics sound classification method based on a nonlinear pattern: Twine-pat

Categorizing the bioacoustic and ecoacoustic properties of animals is great interest to biologists and ecologists. Also, multidisciplinary studies in engineering have significantly contributed to the development of acoustic analysis. Observing the animals living in the ecological environment provides information in many areas such as global warming, climate changes, monitoring of endangered animals, agricultural activities. However, the classification of bioacoustics sounds by manually is very hard. Therefore, automated bioacoustics sound classification is crucial for ecological science. This work presents a new multispecies bioacoustics sound dataset and novel machine learning model to classify bird and anuran species with sounds automatically. In this model, a new nonlinear textural feature generation function is presented by using twine cipher substitution box(S-box), and this feature generation function is named twine-pat. By using twine-pat and tunable Q-factor wavelet transform, a multilevel feature generation network is presented. Iterative ReliefF(IRF) is employed to select the most effective/valuable features. Two shallow classifiers are used to calculate results. Our presented model reached 98.75% accuracy by using k-nearest neighbor(kNN) classifier. The results obviously demonstrated the success of the presented model.     

Geographical variation in the high-duty cycle echolocation of the cryptic common mustached bat Pteronotus cf. rubiginosus (Mormoopidae)

The use of bioacoustics as a tool for bat research is rapidly increasing worldwide. There is substantial evidence that environmental factors such as weather conditions or habitat structure can affect echolocation call structure in bats and thus compromise proper species identification. However, intraspecific differences in echolocation due to geographical variation are poorly understood, which poses a number of issues in terms of method standardization. We examined acoustic data for Pteronotus cf. rubiginosus from the Central Amazon and the Guiana Shield. We provide the first evidence of intraspecific geographic variation in bat echolocation in the Neotropics, with calls significantly differing in almost all standard acoustic parameters for the two lineages of this clade. We complement our bioacoustic data with molecular and morphological data for both species. Considerable overlap in trait values prevents reliable discrimination between the two sympatric Pteronotus based on morphological characters. On the other hand, significant divergence in the frequency of maximum energy suggests that bioacoustics can be used to readily separate both taxa despite extensive intraspecific variability in their echolocation across the Amazon. Given the relative lack of barriers preventing contact between bat populations from the Central Amazon and French Guiana, the documented acoustic variation needs to be further studied in geographically intermediate locations to understand the potential isolation processes that could be causing the described divergence in echolocation and to determine whether this variation is either discrete or continuous.     

Weather conditions determine attenuation and speed of sound: Environmental limitations for monitoring and analyzing bat echolocation

Echolocating bats are regularly studied to investigate auditory‐guided behaviors and as important bioindicators. Bioacoustic monitoring methods based on echolocation calls are increasingly used for risk assessment and to ultimately inform conservation strategies for bats. As echolocation calls transmit through the air at the speed of sound, they undergo changes due to atmospheric and geometric attenuation. Both the speed of sound and atmospheric attenuation, however, are variable and determined by weather conditions, particularly temperature and relative humidity. Changing weather conditions thus cause variation in analyzed call parameters, limiting our ability to detect, and correctly analyze bat calls. Here, I use real‐world weather data to exemplify the effect of varying weather conditions on the acoustic properties of air. I then present atmospheric attenuation and speed of sound for the global range of weather conditions and bat call frequencies to show their relative effects. Atmospheric attenuation is a nonlinear function of call frequency, temperature, relative humidity, and atmospheric pressure. While atmospheric attenuation is strongly positively correlated with call frequency, it is also significantly influenced by temperature and relative humidity in a complex nonlinear fashion. Variable weather conditions thus result in variable and unknown effects on the recorded call, affecting estimates of call frequency and intensity, particularly for high frequencies. Weather‐induced variation in speed of sound reaches up to about ±3%, but is generally much smaller and only relevant for acoustic localization methods of bats. The frequency‐ and weather‐dependent variation in atmospheric attenuation has a threefold effect on bioacoustic monitoring of bats: It limits our capability (1) to monitor bats equally across time, space, and species, (2) to correctly measure frequency parameters of bat echolocation calls, particularly for high frequencies, and (3) to correctly identify bat species in species‐rich assemblies or for sympatric species with similar call designs.     

Testing the performances of automated identification of bat echolocation calls: A request for prudence

Echolocating bats are surveyed and studied acoustically with bat detectors routinely and worldwide, yet identification of species from calls often remains ambiguous or impossible due to intraspecific call variation and/or interspecific overlap in call design. To overcome such difficulties and to reduce workload, automated classifiers of echolocation calls have become popular, but their performance has not been tested sufficiently in the field. We examined the absolute performance of two commercially available programs (SonoChiro and Kaleidoscope) and one freeware package (BatClassify). We recorded noise from rain and calls of seven common bat species with Pettersson real-time full spectrum detectors in Sweden. The programs could always (100%) distinguish rain from bat calls, usually (68–100%) identify bats to group (Nyctalus/Vespertilio/Eptesicus, Pipistrellus, Myotis, Plecotus, Barbastella) and usually (83–99%) recognize typical calls of some species whose echolocation pulses are structurally distinct (Pipistrellus pygmaeus, Barbastella barbastellus). Species with less characteristic echolocation calls were not identified reliably, including Vespertilio murinus (16–26%), Myotis spp. (4–93%) and Plecotus auritus (0–89%). All programs showed major although different shortcomings and the often poor performance raising serious concerns about the use of automated classifiers for identification to species level in research and surveys. We highlight the importance of validating output from automated classifiers, and restricting their use to specific situations where identification can be made with high confidence. For comparison we also present the result of a manual identification test on a random subset of the files used to test the programs. It showed a higher classification success but performances were still low for more problematic taxa.     

A novel approach to field identification of cryptic Apodemus wood mice: calls differ more than morphology

Field identification of European wood mice Apodemus spp. is challenging due to their morphological resemblance and frequent sympatry. We developed discriminant functions based on body mass and acoustic variables of distress calls to identify three cryptic species of wood mice (Apodemus alpicola, Apodemus flavicollis and Apodemus sylvaticus) in Italy. We achieved an overall correct classification rate of 86–98%; the best results (100% correct classification) were obtained for Apodemus sylvaticus calls. This minimally invasive, effective and low‐cost method highlights the potential role of bioacoustics as a powerful tool for field discrimination of cryptic species of terrestrial mammals.     

中国43种蛙类鸣声特征数据集

无尾两栖动物的鸣声通常具有物种特异性,了解其鸣声特征信息,是利用生物声学进行物种多样性调查及物种监测的前提。本文汇总、整理了2012–2020年间利用高保真录音设备在野外记录的43种(隶属于7科26属)无尾两栖动物的鸣声数据,以及相应的鸣声采集信息。对音频文件进行降噪处理后,提供了由61个鸣声的波形图及语图组成的鸣声特征数据集。本数据集展示了鸣声的多种时域和频域信息,如单音节或多音节、音节数、音节时长、音节间隔、鸣声时长、主频、基频、谐波等,为我国无尾两栖类的声学研究、物种多样性调查及鸣声监测提供了数据支持。     

Sex recognition in brown skuas: do acoustic signals matter?

Bird vocalisations are often essential for sex recognition, especially in species that show little morphological sex dimorphism. Brown skuas (Catharacta antarctica lonnbergi), which exhibit uniform plumage across both sexes, emit three main calls: the long call, the alarm call and the contact call. We tested the potential for sex recognition in brown skua calls of 42 genetically sexed individuals by analysing 8–12 acoustic parameters in the temporal and frequency domains of each call type. For every call type, we failed to find sex differences in any of the acoustic parameters measured. Stepwise discriminant function analysis (DFA) revealed that sexes cannot be unambiguously classified, with increasing uncertainty of correct classification from contact calls to long calls to alarm calls. Consequently, acoustic signalling is probably not the key mechanism for sex recognition in brown skuas.     

Passive acoustic monitoring of animal populations with transfer learning

Progress in deep learning, more specifically in using convolutional neural networks (CNNs) for the creation of classification models, has been tremendous in recent years. Within bioacoustics research, there has been a large number of recent studies that use CNNs. Designing CNN architectures from scratch is non-trivial and requires knowledge of machine learning. Furthermore, hyper-parameter tuning associated with CNNs is extremely time consuming and requires expensive hardware. In this paper we assess whether it is possible to build good bioacoustic classifiers by adapting and re-using existing CNNs pre-trained on the ImageNet dataset – instead of designing them from scratch, a strategy known as transfer learning that has proved highly successful in other domains. This study is a first attempt to conduct a large-scale investigation on how transfer learning can be used for passive acoustic monitoring (PAM), to simplify the implementation of CNNs and the design decisions when creating them, and to remove time consuming hyper-parameter tuning phases. We compare 12 modern CNN architectures across 4 passive acoustic datasets that target calls of the Hainan gibbon Nomascus hainanus, the critically endangered black-and-white ruffed lemur Varecia variegata, the vulnerable Thyolo alethe Chamaetylas choloensis, and the Pin-tailed whydah Vidua macroura. We focus our work on data scarcity issues by training PAM binary classification models very small datasets, with as few as 25 verified examples. Our findings reveal that transfer learning can result in up to 82% F1 score while keeping CNN implementation details to a minimum, thus rendering this approach accessible, easier to design, and speeding up further vocalisation annotations to create PAM robust models.     

Identification of sympatric bat species by the echolocation calls

One hundred and thirty-eight echolocation calls of 63 free-flying individuals of five bat species (Rhinolophus ferrumequinum,Myotis formosus,Myotis ikonnikovi,Myotis daubentoni and Murina leucogaster)were recorded (by ultrasonic bat detector (D980)) in Zhi'an village of Jilin Province,China.According to the frequency-time spectra,these calls were categorized into two types:FM/CF (constant frequency) / FM (R.ferrumequinum) and FM (frequency modulated)(M.formosus,M.ikonnikovi,M.daubentoni and M.leucogaster).Sonograms of the calls of R.ferrumequinum could easily be distinguished from those of the other four species.For the calls of the remaining four species,six echolocation call parameters,including starting frequency,ending frequency,peak frequency duration,longest inter-pulse interval and shortest inter-pulse interval,were examined by stepwise discriminant analysis.The results show that 84.1% of calls were correctly classified,which indicates that these parameters of echolocation calls play an important role in identifying bat species.These parameters can be used to test the accuracy of general predictions based on bats' morphology in the same forest and can provide essential information for assessing patterns of bat habitat use.     

Decoding the ‘zeep’ complex: quantitative analysis of interspecific variation in the nocturnal flight calls of nine wood warbler species (Parulidae spp.)

ABSTRACT

The “zeep” complex consists of nine birds that produce nocturnal flight calls with similar acoustic features. Our inability to distinguish these calls inhibits the acoustic monitoring of these species. We test the hypothesis that flight calls of nine warblers in the “zeep” complex show sufficient acoustic differences to allow differentiation. We investigate divergence in these vocalizations by recording birds held for banding and collecting additional recordings from sound libraries. We used three approaches to compare calls between species: analysis of variance in acoustic properties, discriminant analysis of acoustic properties, and spectrographic cross-correlation. The first approach revealed five species that were different in one or more acoustic properties. The second approach revealed a level of assignment to the correct species (73%) that exceeded levels expected by chance (36%). The third approach revealed calls of seven species to be significantly more similar to conspecific calls than heterospecific calls. Our results suggest the calls of many members of the “zeep” complex exhibit species-specific differences in structure, which may allow differentiation of at least five “zeep” species based on call alone. We advocate for the combined use of these three approaches for the comparison of “zeep” calls in future flight call studies.