Frants Havmand Jensen

Understanding which animal produces a sound is essential for animal communication studies and for estimating cue rates for passive acoustic monitoring. In marine environments, biologging tags are used to study vocal behaviour, but efficient underwater sound transmission makes it challenging to identify whether the call is produced by the tagged (focal) animal. In low-frequency baleen whales, animal-borne accelerometers on acoustic recording tags have recorded signals corresponding to vocalizations. Here, we test whether high sample rate accelerometers could estimate focal status with validation datasets in which all members of humpback whale feeding groups were tagged concurrently. Accelerometer signals were only rarely associated with known calls from the tagged animal (3%). In addition, high-amplitude calls from nearby conspecifics were occasionally associated with accelerometer signals, including nine calls present on the accelerometer of two tags simultaneously. Calls were present on the accelerometer only when the acoustic received level was high. Our results indicate that source level and accelerometer noise from movement limit the utility of identifying focal calls from accelerometers, but also highlight the risk of false detections for high-amplitude calls. Group-level tagging provides an alternative way to differentiate calls and investigate communication dynamics in baleen whale social groups.

Most dolphin species produce broadband biosonar clicks for echolocating prey, and whistles and burst‐pulsed calls for communication purposes. Seven dolphin species were previously reported to only produce clicks of a narrowband high‐frequency (NBHF) type, very similar to the clicks produced by porpoises. In this study, we used underwater acoustic recordings of Chilean dolphins ( Cephalorhynchus eutropia ) in southern Chile to show that the click repertoire comprises highly stereotypical NBHF clicks, more variable clicks with lower centroid frequencies and wider bandwidth, and some that are intermediate between these two categories. Both NBHF and broadband clicks were produced in trains, buzzes, and burst‐pulses. Broadband clicks co‐occurred with NBHF clicks in nine out of thirteen Chilean dolphin acoustic encounters, typically during social encounters. We conclude that C. eutropia can relax acoustic crypsis in a similar manner as other members in the Cephalorhynchus genus. Our findings make it likely that other NBHF delphinids have the ability to generate click types with frequencies below the typical NBHF range.

Reliable extraction of dolphin whistle contours is fundamental for scalable analyses of vocal identity, repertoire structure, and individual‐based acoustic monitoring. We evaluated four algorithms for fundamental frequency estimation of common bottlenose dolphin ( Tursiops truncatus ) whistles, leveraging a benchmark dataset of annotated whistles from known animals to assess performance across individuals and as a function of signal‐to‐noise ratio (SNR). In high‐SNR whistles (> 20 dB), the deep learning models CREPE‐tt, SAM‐whistle, and Silbido Profundo achieved similar performance in mean contour coverage (~80%), with CREPE‐tt producing the most continuous contours and the lowest frequency error. SAM‐whistle and Silbido Profundo were more robust than CREPE‐tt under low‐SNR conditions. However, automatically extracted contours consistently reduced within‐ versus between‐individual separation and ultimately decreased individual classification accuracy, highlighting how contour gaps and fragmentation propagate into downstream identity matching. Taken together, our results suggest a practical division: CREPE‐tt excels with high‐SNR data and detailed contour shape analyses without harmonic post‐processing, whereas SAM‐whistle and Silbido Profundo perform better with low‐SNR recordings. For individual identification and abundance estimation, explicit SNR‐based inclusion thresholds and quality‐control criteria remain necessary.

Our ability to study the fine‐scale behavior of cetaceans is evolving rapidly with the use of drones (or unoccupied aircraft systems) as platforms to observe and record behavior and with the emergence of machine learning and data science tools to quantify these observations. Together, these tools can provide an unprecedented perspective on the surface and near‐surface behaviors of cetaceans and the dynamics of their social groups. However, the use of machine learning to study cetacean behavior lags behind its application in other areas, such as cetacean acoustics and photo‐identification. In this review, we discuss the challenges and opportunities of combining these technologies to quantify cetacean social dynamics, and we present recent advances in the development of machine learning pipelines that go beyond the detection of animals. We also consider strategies that can improve the performance of models in individual studies, and as a broader research community, to allow the development of more effective and more generalizable tools.

Here, we present HyenaSET, a large (~1640 hours) bioacoustic dataset derived from collar-mounted audio recorders deployed on 19 spotted hyenas in the Maasai Mara National Reserve, Kenya. Within this dataset, 243 hours have been strongly labeled by identifying the onset and offset of all vocalizations as well as their types, and the labels have been validated by experts on hyena vocalizations and behavior. Within the strongly labeled data, the total amount of time hyenas were vocalizing was 9.5 hours (3.9%). Furthermore, each vocalization has been manually identified as "focal" (emitted by the hyena wearing the collar) or "non-focal" (emitted by a nearby conspecific), making use of information from collar-mounted accelerometers that picked up vibrations of the animal's throat when it produced vocalizations. In addition to the labeled data, we also provide a large corpus of unlabeled data from the same recordings, which can be used for un- or self-supervised machine learning tasks. To ensure reproducibility of this dataset as a benchmark in machine learning studies, we present it alongside five stratified cross-validation train/test splits to enable accurate comparisons, and we also provide a train/test split in which specific individuals are left out of the training set to assess generalizability across individuals. Finally, as a performance benchmark, we present baseline results for this dataset using animal2vec, a recently developed transformer-based model optimized for bioacoustic data.

Anthropogenic noise is considered a marine pollutant by the European Union and can impact marine animals in various ways. Noise that overlaps in time and frequency with animal communication signals can interfere with communication and thus impact ecological functions mediated by acoustic signals, depending on how animals compensate for increasing noise. Our study investigated Lombard responses in long-finned pilot whales using sound and movement recording tags in an area with high anthropogenic pressure, the Strait of Gibraltar. We tested if pilot whales increase their vocal amplitude in response to noise, and whether vocal compensation is signal or context specific. We analyzed 1336 calls from 18 individuals in small social groups of simultaneously tagged animals, allowing us to parse which individual emitted which call and how call output level depended on masking noise level. We found that pilot whales increase their call output level with 0.50±0.04 dB per 1 dB increase in noise level to partially compensate for the effects of masking noise. Call type significantly influenced the Lombard response magnitude, with vocal compensation ranging from 0.11 to 0.87 dB per 1 dB increase in noise level, whereas animals exhibited similar Lombard responses for different dive contexts (non-foraging, deep dive descent and ascent). Our results indicate that pilot whales only compensate partially for increased noise levels, and that Lombard response magnitude depends on signal type. This emphasizes the need for understanding the ecological consequences of communication disruptions to ensure sustainable management of human activities.

Cetaceans are often assumed to employ very high oxygen extractions of ∼40–60% and high tidal volumes (60–80% of vital capacity) to decrease surface time and increase foraging time at depth. However, such oxygen extractions and tidal volumes are greatly at odds with gas exchange in terrestrial mammals, and may, if incorrect, lead to severe overestimations of field metabolic rate (FMR) in wild animals when modeling oxygen uptake from respiration rates. Here, we tested the hypothesis that bottlenose dolphins have such high average oxygen extractions and tidal volumes. By measuring oxygen extractions and tidal volumes of >2000 breaths before and after a 2 min apnea bout in three trained bottlenose dolphins, we show that average pre-apnea resting oxygen extractions are between 17% and 25%, less than half of what has historically been reported for cetaceans. Following apnea, initial oxygen extractions are high (∼60%) but drop below pre-apnea levels in 11–20 breaths. Tidal volumes in this experimental setting were between 21% and 37% of vital capacity, consistent with recent findings for marine mammals, but less than half the 60–80% often assumed for cetaceans in FMR modeling. We therefore reject the hypothesis that bottlenose dolphins on average employ high oxygen extractions and high tidal volumes at rest and following short apneas. Consequently, using fixed high values for tidal volumes and oxygen extractions when modeling FMR from breathing rates in wild cetaceans may possibly lead to overestimations of their energy expenditure, food requirements and ecological roles.

Understanding cetacean echolocation behavior is important for effective population monitoring and conservation. Using passive acoustic monitoring (PAM), researchers can listen for the biosonar clicks produced by echolocating animals to estimate both diurnal and seasonal variations in their presence and activity. Furthermore, if species-specific click rates are known, cue counting techniques can be used to provide an estimate of population density. This study investigated the click rates of wild bottlenose dolphins tagged with sound and movement recording DTAG3s during health assessments over the West Florida Shelf in the Gulf of Mexico to quantify individual variability and explore factors influencing click production. We observed modest but significant differences in click rates across individuals, and higher click rates during dives compared to inter-dive surface intervals. Within dives, dive depth was the most important in shaping click rates, reflecting that dolphins adjust their echolocation behavior to tailor their acoustic field of view based on both predator-prey distance and their proximity to other large reflectors such as the ocean bottom. Click rates also showed subtle diurnal peaks at dawn and dusk, aligning with increased foraging efforts. The findings lay the groundwork for bottlenose dolphin density estimation using the cue counting technique and underscore the importance of incorporating region-specific information on foraging ecology and diving behavior into models of click rates. Our study provides the first estimate of bottlenose dolphin click rates but calls for further research to refine these click rate estimates to facilitate acoustic monitoring of delphinids.

We have built a unique library of sounds produced by known individual common bottlenose dolphins ( Tursiops truncatus ), by recording them non-invasively with suction cup hydrophones during brief catch and release health assessments and with digital acoustic tags (DTAGs). We have catalogued the name-like signature whistles (SWs) of most animals in this resident community of 170 dolphins, which has enabled us to begin studying little known “non-signature whistles” (NSW). We have so far identified 22 shared NSW types, of which two, NSWA and NSWB, are known to have been produced by at least 25 and 35 different dolphins respectively. We are studying the functions of shared NSWs with playback experiments to free-swimming dolphins. We provide background on past playback studies and how they have informed our current research; in particular, received level (RL) of playbacks was found to significantly influence strength of response. Varied responses to playbacks reflect the complexity of dolphin communication, and highlight the need for larger sample sizes to be able to correctly interpret NSW functions. However, results so far have provided support for both the referential nature of SW and the affiliative nature of SW copies (SWCs), because a majority of control playbacks of a dolphin’s own signature whistle (self playbacks) elicited positive responses. NSWA elicited a majority of negative responses, suggesting an alarm-type function, and NSWB elicited varying responses, supporting our suggested function of this whistle type as a “query,” produced when something unexpected or unfamiliar is heard. Given that SW and SWC are known to be learned and appear to be referential signals, it is likely that shared, stereotyped NSW are both learned and referential as well, an idea that is supported by the fact that dolphins are flexible, life-long vocal production learners, unlike most other non-human mammals. Our study provides the first evidence in dolphins for a wider repertoire of shared, context-specific signals, which could form the basis for a language-like communication system.