Mathematical models of long term evolution of blue whale song types' frequencies.

The linear decrease in the frequency of blue whale songs around the world is, to date, an unexplained phenomenon. We show it can be reproduced by a mathematical model considering two antagonistic behavioral trends: first, a bias towards conformity in the song, and second, a tendency to try and sing lower than the other whales. We check the robustness of our model by considering some more complex premises. First, different hierarchical relations between the singers are explored, adapting methods used in the flocking motion studies. Then a population-dependant simulation shows that even considering the gradual addition of new whales, the evolution is still globally linear. Finally, we show that intra-annual variations surging from different causes can be naturally incorporated into the model. We then conclude that, unlike other explanations, a cultural hypothesis seems compatible with the observed linearity of the blue whales's songs frequency shift.

Using context to train time-domain echolocation click detectors.

This work demonstrates the effectiveness of using humans in the loop processes for constructing large training sets for machine learning tasks. A corpus of over 57 000 toothed whale echolocation clicks was developed by using a permissive energy-based echolocation detector followed by a machine-assisted quality control process that exploits contextual cues. Subsets of these data were used to train feed forward neural networks that detected over 850 000 echolocation clicks that were validated using the same quality control process. It is shown that this network architecture performs well in a variety of contexts and is evaluated against a withheld data set that was collected nearly five years apart from the development data at a location over 600 km distant. The system was capable of finding echolocation bouts that were missed by human analysts, and the patterns of error in the classifier consist primarily of anthropogenic sources that were not included as counter-training examples. In the absence of such events, typical false positive rates are under ten events per hour even at low thresholds.

A standardized method of classifying pulsed sounds and its application to pulse rate measurement of blue whale southeast Pacific song units.

Pulsed sounds are an interesting example of complex biological sounds, frequent in cetaceans' vocalizations. A two-group classification of these sounds is proposed: tonal or non-tonal. Through two simple mathematical models, it is shown that this classification can lead to better techniques for measuring the pulse rate. This classification is thus useful for improving measurement accuracy, but can also help in formulating hypotheses regarding mechanisms of sound production. This method of classification is applied to south Pacific blue whale vocalizations and it is found that the pulse rate corresponds to the fundamental frequency (not expressed in the spectrum) of the song. Thus, the hypothesis that the sound is produced by only one organ and then filtered by the body of the giant is reinforced.

Classification of mysticete sounds using machine learning techniques.

Automatic classification of mysticete sounds has long been a challenging task in the bioacoustics field. The unknown statistical properties of the signals as well as the use of different recording apparatus and low signal-to-noise ratio conditions often lead to non-optimal systems. The goal of this paper is to design methods for the automatic classification of mysticete sounds using a restricted Boltzmann machine and a sparse auto-encoder that are widely used in the field of artificial intelligence. Experiments on five species of mysticetes are presented. The different methods are employed on the subset of species whose frequency range overlaps, as well as in all five species' calls. Moreover, results are offered with and without the use of a noise class. Overall, the systems are able to achieve an average classification accuracy of over 69% (with noise) and 80% (without noise) given the different architectures.

Deep audio embeddings for vocalisation clustering.

The study of non-human animals' communication systems generally relies on the transcription of vocal sequences using a finite set of discrete units. This set is referred to as a vocal repertoire, which is specific to a species or a sub-group of a species. When conducted by human experts, the formal description of vocal repertoires can be laborious and/or biased. This motivates computerised assistance for this procedure, for which machine learning algorithms represent a good opportunity. Unsupervised clustering algorithms are suited for grouping close points together, provided a relevant representation. This paper therefore studies a new method for encoding vocalisations, allowing for automatic clustering to alleviate vocal repertoire characterisation. Borrowing from deep representation learning, we use a convolutional auto-encoder network to learn an abstract representation of vocalisations. We report on the quality of the learnt representation, as well as of state of the art methods, by quantifying their agreement with expert labelled vocalisation types from 8 datasets of other studies across 6 species (birds and marine mammals). With this benchmark, we demonstrate that using auto-encoders improves the relevance of vocalisation representation which serves repertoire characterisation using a very limited number of settings. We also publish a Python package for the bioacoustic community to train their own vocalisation auto-encoders or use a pretrained encoder to browse vocal repertoires and ease unit wise annotation.

Medium-term acoustic monitoring of small cetaceans in Patagonia, Chile.

Coastal dolphins and porpoises such as the Chilean dolphin (Cephalorhynchus eutropia), the Peale's dolphin (Lagenorhynchus australis), and the Burmeister's porpoise (Phocoena spinipinnis) inhabit the remote areas of Chilean Patagonia. Human development is growing fast in these parts and may constitute a serious threat to such poorly known species. It is thus urgent to develop new tools to try and study these cryptic species and find out more about their behavior, population levels, and habits. These odontocetes emit narrow-band high-frequency (NBHF) clicks and efforts have been made to characterize precisely their acoustic production. Passive acoustic monitoring is a common way to study these animals. Nevertheless, as the signal frequency is usually higher than 100 kHz, storage problems are acute and do not allow for long-term monitoring. The solutions for recording NBHF clicks are usually twofold: either short duration, opportunistic recording from a small boat in presence of the animals (short-term monitoring) or long-term monitoring using devices including a click detector and registering events rather than sound. We suggest, as another possibility, medium-term monitoring, arguing that today's devices have reached a level of performance allowing for a few days of continual recording even at these extremely high frequencies and in difficult conditions, combined with a long-term click detector. As an example, during 2021, we performed a quasi-continuous recording for one week with the Qualilife High-Blue recorder anchored in a fjord near Puerto Cisnes, Region de Aysen, Chile. We detected more than 13,000 clicks, grouped in 22 periods of passing animals. Our detected clicks are quite similar to precedent results but, due to the large number of clicks recorded, we find a larger variability of parameters. Several rapid sequences of clicks (buzz) were found in the recordings and their features are consistent with previous studies: on average they have a larger bandwidth and a lower peak frequency than the usual clicks. We also installed in the same place a click detector (C-POD) and the two devices compare well and show the same number and duration of periods of animals presence. Passages of odontocetes were happening on average each three hours. We thus confirm the high site fidelity for the species of dolphins emitting NBHF clicks present in this zone. Finally, we confirm that the combined use of recording and detection devices is probably a good alternative to study these poorly known species in remote areas.

Passive acoustic monitoring of sperm whales and anthropogenic noise using stereophonic recordings in the Mediterranean Sea, North West Pelagos Sanctuary.

A total of 147 days spread over 4 years were recorded by a stereophonic sonobuoy set up in the Mediterranean sea, near the coast of Toulon, south of France. These recordings were analyzed in the scope of studying sperm whales (Physeter macrocephalus) and the impact anthropic noises may have on this species. With the use of a novel approach, which combines the use of a stereophonic antenna with a neural network, 226 sperm whales' passages have been automatically detected in an effective range of 32 km. This dataset was then used to analyze the sperm whales' abundance, the background noise, the influence of the background noise on the acoustic presence, and the animals' size. The results show that sperm whales are present all year round in groups of 1-9 individuals, especially during the daytime. The estimated density is 1.69 whales/1000 km[Formula: see text]. Animals were also less frequent during periods with an increased background noise due to ferries. The animal size distribution revealed the recorded sperm whales were distributed in length from about 7 to 15.5 m, and lonely whales are larger, while groups of two are composed of juvenile and mid-sized animals.

Temporal evolution of the Mediterranean fin whale song.

We present an analysis of fin whale (Balaenoptera physalus) songs on passive acoustic recordings from the Pelagos Sanctuary (Western Mediterranean Basin). The recordings were gathered between 2008 and 2018 using 2 different hydrophone stations. We show how 20 Hz fin whale pulses can be automatically detected using a low complexity convolutional neural network (CNN) despite data variability (different recording devices exposed to diverse noises). The pulses were further classified into the two categories described in past studies and inter pulse intervals (IPI) were measured. The results confirm previous observations on the local relationship between pulse type and IPI with substantially more data. Furthermore we show inter-annual shifts in IPI and an intra-annual trend in pulse center frequency. This study provides new elements of comparison for the understanding of long term fin whale song trends worldwide.

Kin relationships in cultural species of the marine realm: case study of a matrilineal social group of sperm whales off Mauritius island, Indian Ocean.

Understanding the organization and dynamics of social groups of marine mammals through the study of kin relationships is particularly challenging. Here, we studied a stable social group of sperm whales off Mauritius, using underwater observations, individual-specific identification, non-invasive sampling and genetic analyses based on mitochondrial sequencing and microsatellite profiling. Twenty-four sperm whales were sampled between 2017 and 2019. All individuals except one adult female shared the same mitochondrial DNA (mtDNA) haplotype-one that is rare in the western Indian Ocean-thus confirming with near certainty the matrilineality of the group. All probable first- and second-degree kin relationships were depicted in the sperm whale social group: 13 first-degree and 27 second-degree relationships were identified. Notably, we highlight the likely case of an unrelated female having been integrated into a social unit, in that she presented a distinct mtDNA haplotype and no close relationships with any members of the group. Investigating the possible matrilineality of sperm whale cultural units (i.e. vocal clans) is the next step in our research programme to elucidate and better apprehend the complex organization of sperm whale social groups.

Inter-annual decrease in pulse rate and peak frequency of Southeast Pacific blue whale song types.

A decrease in the frequency of two southeast Pacific blue whale song types was examined over decades, using acoustic data from several different sources in the eastern Pacific Ocean ranging between the Equator and Chilean Patagonia. The pulse rate of the song units as well as their peak frequency were measured using two different methods (summed auto-correlation and Fourier transform). The sources of error associated with each measurement were assessed. There was a linear decline in both parameters for the more common song type (southeast Pacific song type n.2) between 1997 to 2017. An abbreviated analysis, also showed a frequency decline in the scarcer southeast Pacific song type n.1 between 1970 to 2014, revealing that both song types are declining at similar rates. We discussed the use of measuring both pulse rate and peak frequency to examine the frequency decline. Finally, a comparison of the rates of frequency decline with other song types reported in the literature and a discussion on the reasons of the frequency shift are presented.

NIPS4Bplus: a richly annotated birdsong audio dataset.

Recent advances in birdsong detection and classification have approached a limit due to the lack of fully annotated recordings. In this paper, we present NIPS4Bplus, the first richly annotated birdsong audio dataset, that is comprised of recordings containing bird vocalisations along with their active species tags plus the temporal annotations acquired for them. Statistical information about the recordings, their species specific tags and their temporal annotations are presented along with example uses. NIPS4Bplus could be used in various ecoacoustic tasks, such as training models for bird population monitoring, species classification, birdsong vocalisation detection and classification.

Behavioural responses of humpback whales to food-related chemical stimuli.

Baleen whales face the challenge of finding patchily distributed food in the open ocean. Their relatively well-developed olfactory structures suggest that they could identify the specific odours given off by planktonic prey such as krill aggregations. Like other marine predators, they may also detect dimethyl sulfide (DMS), a chemical released in areas of high marine productivity. However, dedicated behavioural studies still have to be conducted in baleen whales in order to confirm the involvement of chemoreception in their feeding ecology. We implemented 56 behavioural response experiments in humpback whales using two food-related chemical stimuli, krill extract and DMS, as well as their respective controls (orange clay and vegetable oil) in their breeding (Madagascar) and feeding grounds (Iceland and Antarctic Peninsula). The whales approached the stimulus area and stayed longer in the trial zone during krill extract trials compared to control trials, suggesting that they were attracted to the chemical source and spent time exploring its surroundings, probably in search of prey. This response was observed in Iceland, and to a lesser extend in Madagascar, but not in Antarctica. Surface behaviours indicative of sensory exploration, such as diving under the stimulus area and stopping navigation, were also observed more often during krill extract trials than during control trials. Exposure to DMS did not elicit such exploration behaviours in any of the study areas. However, acoustic analyses suggest that DMS and krill extract both modified the whales' acoustic activity in Madagascar. Altogether, these results provide the first behavioural evidence that baleen whales actually perceive prey-derived chemical cues over distances of several hundred metres. Chemoreception, especially olfaction, could thus be used for locating prey aggregations and for navigation at sea, as it has been shown in other marine predators including seabirds.

Bottlenose dolphin (Tursiops truncatus) sonar slacks off before touching a non-alimentary target.

Odontocetes modulate the rhythm of their echolocation clicks to draw information about their environment. When they approach preys to capture, they speed up their emissions to increase the sampling rate of "distant touch" and improve information update. This global acceleration turns into a "terminal buzz" also described in bats, which is a click train with drastic increase in rate, just as reaching the prey. This study documents and analyses under human care bottlenose dolphins' echolocation activity, when approaching non-alimentary targets. Four dolphins' locomotor and clicking behaviours were recorded during training sessions, when sent to immersed objects pointed by their trainers. Results illustrate that these dolphins profusely use echolocation towards immersed non-alimentary objects. They accelerate click emission when approaching the target, thus displaying a classical terminal buzz. However, their terminal buzz slackens off within a quarter of second before the end of click train. Typically, they decelerate to stop clicking just before they touch the object using their rostrum lower tip. They do not emit clicks as the contact lasts. In conclusion, when exploring inert objects, bottlenose dolphins under human accelerate clicking like other odontocetes or bats approaching preys. Bottlenose dolphins' particular slackening-off profile at the end of the buzz shows that they anticipate the moment of direct contact, and they stop just as real touch relays distant touch of the object.

Cooperative sparse representation in two opposite directions for semi-supervised image annotation.

Recent studies have shown that sparse representation (SR) can deal well with many computer vision problems, and its kernel version has powerful classification capability. In this paper, we address the application of a cooperative SR in semi-supervised image annotation which can increase the amount of labeled images for further use in training image classifiers. Given a set of labeled (training) images and a set of unlabeled (test) images, the usual SR method, which we call forward SR, is used to represent each unlabeled image with several labeled ones, and then to annotate the unlabeled image according to the annotations of these labeled ones. However, to the best of our knowledge, the SR method in an opposite direction, that we call backward SR to represent each labeled image with several unlabeled images and then to annotate any unlabeled image according to the annotations of the labeled images which the unlabeled image is selected by the backward SR to represent, has not been addressed so far. In this paper, we explore how much the backward SR can contribute to image annotation, and be complementary to the forward SR. The co-training, which has been proved to be a semi-supervised method improving each other only if two classifiers are relatively independent, is then adopted to testify this complementary nature between two SRs in opposite directions. Finally, the co-training of two SRs in kernel space builds a cooperative kernel sparse representation (Co-KSR) method for image annotation. Experimental results and analyses show that two KSRs in opposite directions are complementary, and Co-KSR improves considerably over either of them with an image annotation performance better than other state-of-the-art semi-supervised classifiers such as transductive support vector machine, local and global consistency, and Gaussian fields and harmonic functions. Comparative experiments with a nonsparse solution are also performed to show that the sparsity plays an important role in the cooperation of image representations in two opposite directions. This paper extends the application of SR in image annotation and retrieval.