Passive acoustic monitoring of fish choruses: a review to inform the development of a monitoring and management tool

Reviews in Fish Biology and Fisheries, Год журнала: 2025, Номер unknown

Опубликована: Март 19, 2025

Язык: Английский

---

Applications of machine learning to identify and characterize the sounds produced by fish

ICES Journal of Marine Science, Год журнала: 2023, Номер 80(7), С. 1854 - 1867

Опубликована: Авг. 11, 2023

Abstract Aquatic ecosystems are constantly changing due to anthropic stressors, which can lead biodiversity loss. Ocean sound is considered an essential ocean variable, with the potential improve our understanding of its impact on marine life. Fish produce a variety sounds and their choruses often dominate underwater soundscapes. These have been used assess communication, behaviour, spawning location, biodiversity. Artificial intelligence provide robust solution detect classify fish sounds. However, main challenge in applying artificial recognize lack validated data for individual species. This review provides overview recent publications use machine learning, including deep detection, classification, identification. Key challenges limitations discussed, some points guide future studies also provided.

Язык: Английский

---

Automatic detection of unidentified fish sounds: a comparison of traditional machine learning with deep learning

Frontiers in Remote Sensing, Год журнала: 2024, Номер 5

Опубликована: Авг. 22, 2024

Many species of fishes around the world are soniferous. The types sounds produce vary among and regions but consist typically low-frequency ( < 1.5 kHz) pulses grunts. These can potentially be used to monitor non-intrusively could complement traditional monitoring techniques. However, significant time required for human analysts manually label fish in acoustic recordings does not yet allow passive acoustics as a viable tool fishes. In this paper, we compare two different approaches automatically detect sounds. One is more machine learning technique based on detection transients spectrogram classification using Random Forest (RF). other deep approach overlapping segments (0.2 s) ResNet18 Convolutional Neural Network (CNN). Both algorithms were trained 21,950 annotated non-fish collected from 2014 2019 at five locations Strait Georgia, British Columbia, Canada. performance detectors was tested part data Georgia that withheld training phase, Barkley Sound, Port Miami, Florida, United States. CNN performed up 1.9 times better than RF id="m2">F1 score: 0.82 vs. 0.43). some cases, able find faint analyst well environments one it (Miami id="m3">F1 0.88). Noise analysis 20–1,000 Hz frequency band shows still reliable noise levels greater 130 dB re 1 id="m4">μ Pa Miami becomes less Sound past 100 id="m5">μ due mooring noise. proposed efficiently (unidentified) variety also facilitate development species-specific detectors. We provide software FishSound Finder, an easy-to-use open-source implementation detector with detailed documentation.

Язык: Английский

---

Unlocking the soundscape of coral reefs with artificial intelligence: pretrained networks and unsupervised learning win out

PLoS Computational Biology, Год журнала: 2025, Номер 21(4), С. e1013029 - e1013029

Опубликована: Апрель 28, 2025

Passive acoustic monitoring can offer insights into the state of coral reef ecosystems at low-costs and over extended temporal periods. Comparison whole soundscape properties rapidly deliver broad from data, in contrast to detailed but time-consuming analysis individual bioacoustic events. However, a lack effective automated for data has impeded progress this field. Here, we show that machine learning (ML) be used unlock greater soundscapes. We showcase on diverse set tasks using three biogeographically independent datasets, each containing fish community (high or low), cover low) depth zone (shallow mesophotic) classes. supervised train models identify ecological classes sites report unsupervised clustering achieves whilst providing more understanding site groupings within data. also compare different approaches extracting feature embeddings recordings input ML algorithms: indices commonly by ecologists, pretrained convolutional neural network (P-CNN) trained 5.2 million hrs YouTube audio, CNN’s which were task (T-CNN). Although T-CNN performs marginally better across tasks, reveal P-CNN offers powerful tool generating marine as it requires orders magnitude less computational resources achieving near comparable performance T-CNN, with significant improvements indices. Our findings have implications ecology any habitat.

Язык: Английский

---

Climatic and economic fluctuations revealed by decadal ocean soundscapes

The Journal of the Acoustical Society of America, Год журнала: 2025, Номер 157(6), С. 4233 - 4251

Опубликована: Июнь 1, 2025

Decadal variations of ocean soundscapes are intricately linked to large-scale climatic and economic fluctuations. This study draws on over 15 years acoustic recordings at six sites within the Southern California Bight, investigating interannual, seasonal, diel variations. By examining energy from fin blue whales along with sounds ships wind, we identified changes in soundscape time space. reveals that sound levels associated both biological non-biological sources varied seasonally correlated patterns long-term oceanographic Baleen whale before, during, after a marine heatwave were assessed; decreased southern increased northern adjacent Current, underscoring potential for range shifts habitat compression during warm these species. Ship-generated high-traffic reflected events such as recessions, labor shortages negotiations, port activities. Marine offer an approach assess ocean's condition amid ongoing

Язык: Английский

---

Unlocking the soundscape of coral reefs with artificial intelligence: pretrained networks and unsupervised learning win out

bioRxiv (Cold Spring Harbor Laboratory), Год журнала: 2024, Номер unknown

Опубликована: Фев. 7, 2024

Abstract Passive acoustic monitoring can offer insights into the state of coral reef ecosystems at low-costs and over extended temporal periods. Comparison whole soundscape properties rapidly deliver broad from data, in contrast to more detailed but time-consuming analysis individual bioacoustic signals. However, a lack effective automated for data has impeded progress this field. Here, we show that machine learning (ML) be used unlock greater soundscapes. We showcase on diverse set tasks using three biogeographically independent datasets, each containing fish community, cover or depth zone classes. supervised train models identify ecological classes sites report unsupervised clustering achieves whilst providing understanding site groupings within data. also compare different approaches extracting feature embeddings recordings input ML algorithms: indices commonly by ecologists, pretrained convolutional neural network (P-CNN) trained 5.2m hrs YouTube audio CNN datasets (T-CNN). Although T-CNN performs marginally better across reveal P-CNN is powerful tool identifying marine ecologists due its strong performance, low computational cost significantly improved performance indices. Our findings have implications ecology any habitat. Author Summary Artificial intelligence potential revolutionise reefs. So far, limited work detectors specific sounds such as species. building process involves manually annotating large amounts followed complicated model training, must then repeated all again new dataset. Instead, explore techniques analysis, which compares raw entire multiple methods rigorously test these Indonesia, Australia French Polynesia. key use hours unrelated offers produce compressed representations conserving data’s being executable standard personal laptop. These patterns soundscapes “unsupervised learning”, grouping similar periods together dissimilar apart. hold relationships with ground truth including coverage, community depth.

Язык: Английский

---

Fish Acoustic Detection Algorithm Research: a deep learning app for Caribbean grouper calls detection and call types classification

Frontiers in Marine Science, Год журнала: 2024, Номер 11

Опубликована: Июль 22, 2024

In this paper, we present the first machine learning package developed specifically for fish calls identification within a specific range (0–500Hz) that encompasses four Caribbean grouper species: red hind ( E. guttatus ), Nassau striatus yellowfin M. venenosa and black bonaci ). Because of their ubiquity in soundscape grouper’s habitat, squirrelfish Holocentrus spp.) sounds along with vessel noise are also detected. addition model is able to separate species call types. This called FADAR, Fish Acoustic Detection Algorithm Research standalone user-friendly application Matlab™ . The concept FADAR product evaluation various deep architectures have been presented series published articles. composed main algorithm can detect all including architecture based on an ensemble approach where bank five CNNs randomly assigned hyperparameters used form classifiers. outputs combined by fusion process decision making. At level, output multimodel thus classify terms done models thoroughly evaluated literature concerned here, transfer groupers custom designed CNN grouper, which has greater number known types than other species. was manually trained diversity data span regions Sea two recorder brands, hydrophone sensitivities, calibrations sampling rates, mobile platform. strategy conferred substantive robustness level sources be found frequency band such as vessels marine mammals. Performance metrics sensitivity (recall) specificity showed same performance both balanced unbalanced datasets at locations not training set.

Язык: Английский

---

Automated cataloging of oyster toadfish (Opsanus tau) boatwhistle calls using template matching and machine learning

Ecological Informatics, Год журнала: 2023, Номер 77, С. 102268 - 102268

Опубликована: Авг. 22, 2023

Oyster toadfish (Opsanus tau) represent an ecologically significant species found throughout estuaries along the eastern coast of United States. While these crevice-dwelling fish can be challenging to observe in their habitats, it is possible infer distribution and aspects behavior by recording sounds they produce. The task cataloging distinctive advertisement boatwhistle produced male attract females spring summer automated using a multi-step process. Candidate boatwhistles are first identified template matching suite synthetic spectrogram kernels formed mimic two lowest frequency harmonic tones within boatwhistle. calls based on correlation between low-frequency data. Next, frequency-reassigned images candidates input into pre-trained ResNet-50 convolutional neural network. Finally, activations from deep, fully connected layer this network extracted passed one-vs-all support-vector-machine classifier, which separates larger set candidate signals. This classifier model was trained evaluated labeled dataset over 20,000 signals generated diverse acoustic conditions Pamlico Sound, North Carolina, USA. accompanying software provides effective efficient tool monitor calls, may facilitate deeper understanding spatial distribution, behavioral patterns, ecological roles played oyster toadfish.

Язык: Английский

---

Introduction to the special issue on fish bioacoustics: Hearing and sound communication

The Journal of the Acoustical Society of America, Год журнала: 2024, Номер 155(4), С. 2385 - 2391

Опубликована: Апрель 1, 2024

Fish bioacoustics, or the study of fish hearing, sound production, and acoustic communication, was discussed as early Aristotle. However, questions about how fishes hear were not really addressed until 20th century. Work on bioacoustics grew after World War II considerably in 21st century since investigators, regulators, others realized that anthropogenic (human-generated sounds), which had primarily been interest to workers marine mammals, likely have a major impact (as well aquatic invertebrates). Moreover, passive monitoring fishes, recording sounds field, has blossomed noninvasive technique for sampling abundance, distribution, reproduction various sonic fishes. The field is vital invertebrates make up portion protein eaten by signification humans. To help better understand engage it with issues sound, this special issue Journal Acoustical Society America (JASA) brings together papers explore breadth topic, from historical perspective latest findings

Язык: Английский

---

Exploring fish choruses: patterns revealed through PCA computed from daily spectrograms

Frontiers in Antennas and Propagation, Год журнала: 2024, Номер 2

Опубликована: Июль 25, 2024

Soundscape analysis has become integral to environmental monitoring, particularly in marine and terrestrial settings. Fish choruses within ecosystems provide essential descriptors for characterization. This study employed a month-long sequence of continuous underwater recordings generate 24-h spectrograms, utilizing Principal Component Analysis (PCA) specifically adapted analyze fish choruses. The spectrograms were constructed using frequency range from 0 5 kHz, represented by 1,025 spectral points (frequency bin width Hz) on linear scale. A preliminary subsampling reduced the components 205 points. PCA was then applied this subsampled data, selecting 7 principal (PCs) that explained 95% variance. To enhance visualization interpretation, we introduced “acoustic maps” portrayed as heatmaps. methodology proved valuable characterizing structure observed environment capturing pertinent diel patterns Additionally, these can be analyzed acoustic maps reveal hidden dynamics environment. dimensionality reduction achieved not only streamlined data handling but also enabled extraction information temporal soundscape. In conclusion, our presents versatile framework extendable diverse biological ecoacoustic studies. straightforward, easily interpretable leverages computations derived offering novel insights into daily biological. Choruses contributing future advancements research.

Язык: Английский

---