individual sounds or mass phenomena (e.g., fish choruses), acoustic signals provide valuable information about sound-producers, their associated populations, and evolutionary histories (e.g., Rice et al. 2022;Almagro et al. 2024). Biological sounds may also serve as proxies for biodiversity and biotic-abiotic interactions -even if the sound sources are yet to be identified (Desiderà et al. 2019;Hawkins et al. 2025).Despite the growing recognition of the importance and utility of underwater sounds, many soniferous species and their sounds remain to be identified. Only 4% of fish species and even fewer underwater invertebrates and plants have been tested for sound production (Looby et al. 2022(Looby et al. , 2024;;Desjonquères et al. 2024;van der Lee et al. 2025). New reports of sonifery for previously untested species are being documented every year (e.g., Almagro et al. 2024), aided in part by new technological advances (e.g., Mouy et al. 2023;Dantzker et al. 2025), but the rate of discovery remains woefully inadequate to document the many thousands of soniferous species thought to exist (Desjonquères et al. 2024;Looby et al. 2022Looby et al. , 2023)). Similarly, even for relatively well-documented sounds, their sources may take years to identify (e.g., Bolgan et al. 2019;Allen et al. 2024). Consequently, except for marine mammals, the majority of communicative sounds recorded in the biological component of a soundscape remain of unknown origin. This is particularly true in coastal biodiversityhotspots such as coral reefs and seagrass meadows dominated by sounds from invertebrates and fishes (Tricas and Boyle 2014;Di Iorio et al. 2021;Bolgan et al. 2022), as well as in understudied deep-sea ecosystems (Lin and Kawagucci 2024).While research and monitoring applications can be maximized if sound sources are identified to species, even sounds from these unknown sources can still provide ecologically relevant information. Many fish, for instance, produce a variety of sounds that are components of acoustic communities whose diversity and richness are related to taxonomic richness and diversity (Desiderà et al. 2019;Carrico et al. 2020). These acoustic communities of primarily unknown sounds are also habitat-specific, reflect habitat quality, and respond to management actions (Di Iorio et al. 2021;LaManna et al. 2021;Raick et al. 2023). Changes in the composition of these acoustic communities likely reflect changes in behaviours or underlying assemblages related to environmental variations or human impact. As unidentified sounds present unique challenges, however, they are not commonly prioritized and are often underreported-serving as an elusive yet valuable source of ecological and evolutionary knowledge. As the identification of new unknown sounds is constantly increasing-at a much higher pace than the identification of their sources-their description and characterization is critical to establish reference catalogues as well as adequate detection and classification algorithms to capture regime shift and biodiversity dynamics that may have implications for ecological processes and ecosystems.The Research Topic presented here provides an exploration of sound types, patterns, detection advances, and pitfalls, encompassing various habitats, regions, and taxonomic groups. The articles are centered around four primary themes: advancements in automated detection and classification; global baselines and regional catalogues; spatiotemporal and behavioral dynamics; and acoustic indicators of biodiversity and habitat health. Collectively, these papers highlight the value of studying unidentified underwater biological sounds as well as important developments in their documentation.A central challenge in ecoacoustics is the labor-intensive nature of manually annotating acoustic data, which limits the scale of passive acoustic monitoring studies. Mouy et al. (2024) address this by comparing a traditional machine learning approach (i.e., Random Forest) with a deep learning framework (i.e., ResNet18 Convolutional Neural Network) to detect unidentified fish sounds. The Convolutional Neural Network significantly outperformed the Random Forest model, with demonstrated generalizability by effectively detecting sounds in both Canadian and Floridian ecosystems. The study authors also go beyond merely reporting their analysis findings by sharing their open-source Python software, FishSound Finder, to facilitate future monitoring of fish knock and grunt sounds.propose a novel methodology using the YOLOv8 computer vision algorithm to detect salient acoustic events on a spectrogram. By incorporating an active learning approach, the model iteratively selects the most informative audio files for manual annotation, thereby reducing the human effort required to train robust detectors. Once detected, events are clustered using BioLingual embeddings, providing a framework for characterizing soundscapes even when sources remain unknown.Similarly, Calonge et al. ( 2024) explore unsupervised clustering of unknown sounds in a region of the North Sea dominated by anthropogenic noise. The study compares feature extraction methods, including the Animal Vocalization Encoder based on Self-Supervision (AVES), to derive meaningful clusters from human annotations. The resulting 26 clusters provide a systematic baseline for building training datasets in complex, turbid environments where visual surveys are restricted.Recognizing the need for data standardization, Hawkins et al. (2024) present the Australian Fish Chorus Catalogue (AFCC), an inventory of 301 choruses detected across 83 locations. This open-access repository includes spectral measurements, audio examples, and seasonal presence data extracted via reproducible methodologies. The AFCC establishes a national baseline for tracking the impacts of anthropogenic and environmental change on soniferous fish populations. 2024) demonstrate the utility of cross-referencing such catalogues by comparing unidentified fish sounds across the Azores, Madeira, and mainland Portugal. By matching species lists with common sound types, the authors narrowed down potential sources to a few fish families. This cross-regional comparison highlights how growing databases can unravel the origins of mysterious sounds across vast distances.California Bight known as "The Wave," where a biological chorus propagates upcoast at approximately 1.5 km/s. Using cellular automata models based on the physics of excitable media, the authors successfully replicated the cycling nature of the chorus and its sensitivity to external stimuli like ship noise. This modeling framework provides a new avenue for inverting acoustic measurements to estimate the spatial density of vocalizing animals. 2025) challenge the concept of discrete sound types by examining downsweeps from ten mysticete species in Australian waters. The study found that these calls do not form clear clusters but instead exist as a "graded continuum" of acoustic features that morph gradually in duration, frequency, and decoration. Their findings emphasize that assigning calls to species based solely on spectrographic shape is often inconclusive without regional ecological and behavioral context.Jarriel et al. ( 2024) address the challenge of identifying reliable acoustic metrics for coral reef health by investigating how manually audited low-frequency fish call rates relate to traditional visual surveys. The study found that manual call rates successfully differentiated between reefs of varying quality in the U.S. Virgin Islands, serving as a stronger predictor of hard coral cover and fish abundance than common computational indices like the Acoustic Complexity Index (ACI).Expanding on individual call components, Noble et al. ( 2024) use unsupervised machine learning to categorize the fine-scale composition of pulsed calls on a Caribbean reef. The authors identified 55 distinct sound motifs, revealing substantial acoustic diversity and evidence of spectral and temporal niche differentiation among soniferous reef communities. Their findings suggest that fish may stratify the acoustic domain to identify conspecifics and avoid interference.Taking a global perspective, Chapuis et al. ( 2025) analyze over 120,000 fish calls across six disparate bioregions to determine if reef fish sounds exhibit consistent structural traits. Using Geometric Morphometrics (GMM) and Principal Component Analysis (PCA), the study revealed a "universal symphony" of short, low-frequency sounds that remained structurally conserved regardless of geographical origin. This consistency likely reflects shared environmental and biological constraints common to reef habitats worldwide, such as propagation efficiency and conserved anatomical structures.Finally, Havlik et al. ( 2025) extend these ecological assessments to the understudied mesophotic coral ecosystems of the central Red Sea. The study cataloged 11 nocturnal fish choruses across shallow and deep reef zones, documenting significant seasonal shifts in activity. Most choruses were found to originate near mesophotic sites, highlighting these deeper reefs as critical habitats for foraging, courtship, or spawning.Collectively, these contributions present conceptual and methodological shifts in the effort to detect, categorise, and document unidentified underwater biological sounds. They highlight the diversity of research topics and identify gaps. For example, a major common thread was the integration of advanced Automated Detection and Classification Tools to bridge the gap between massive acoustic datasets and human analytical capacity. Compared to computationally simpler acoustic indices (e.g., Acoustic Complexity Index) that may have inconsistent results, the analysis of individual signals and their patterns is increasingly providing an alternative and potentially complementary pathway for deriving ecological insights, offering a promising direction for sound diversity assessment. Many analytical approaches-such as active learning and unsupervised clustering-are emerging to reduce the 'bottleneck' of manual annotation. The transition from 'hand-crafted' features to deep learning embeddings (e.g., convolutional neural networks) represents another foundational methodological advance. However, to more effectively move beyond detection and towards exploring relationships between acoustic diversity and ecology, manual processing and artificial intelligence tools should work 'hand in code', which would be aided through the establishment of Global Baselines and Regional Catalogues. The diversity of unidentified sound types documented in the compiled contributions underscores the necessity for comprehensive, shared catalogues.The diversity of unidentified sounds also reflects the wealth of biological and ecological information that can be derived from biophonies. For example, analyzing Spatiotemporal and Behavioral Dynamics of biological sounds enables researchers to estimate and predict animal abundance and distributions, track the impacts of anthropogenic and environmental changes in the future, and determine the biological or ecological context in which sounds are produced. Evaluating biophonic diversity, comparing sound types across bioregions, and examining the composition of acoustic spaces can yield valuable insights into habitats that are otherwise challenging to study-such as mesophotic reefs-as well as into biogeographical patterns, interspecies interactions within animal communities, and relationships between acoustic communities and their habitats. Integrating such information into analyses may lead to more consistent Acoustic Indicators of Biodiversity and Health and more robust interpretation of their ecological relevance, a critical factor if these indicators are to be used for environmental monitoring and management Despite these advances, several significant gaps remain. The world's waters are vast and overwhelmingly acoustically unexplored, and the application of PAM in the aquatic environment is expanding rapidly across the globe. In terms of unidentified sounds, PAM remains in a time of discovery. New sounds are being uncovered far more quickly than sources are being identified, and the opportunities for natural observations to glean ecological meaning of each sound are much smaller than the volumes of data being remotely collected. Thus, the primary challenge continues to be the lack of 'ground-truthing' to link specific sounds to their biological sources and associated functions. Some sounds, however, may remain difficult to assign to species based on their acoustic characteristics. Even if sources remain obscure, evaluating the ecological relevance of sound types, signal rates, and the effect of environmental and anthropogenic context on the detected signals remains vital to the generalized application of any emerging artificial intelligence techniques and is unlikely to be achieved with artificial intelligence methods alone. Without identification or ecoacoustic contexts, results may remain open to misinterpretation. Additionally, the need for a unifying Global Library of Underwater Biological Sounds (GLUBS; Parsons et al. 2022) remains paramount for future progress. Such a library would allow comparisons between recorded sound types and species-specific or unknown sounds, as well as the establishment of links between sound types across the globe, identifying shared or specific sound types.The ability to describe spatiotemporal changes in biodiversity, detect regime shifts, and assess ecosystem condition, all by simply listening under water-even when the individual callers remain anonymous-is a vital step for non-invasive monitoring, management, and conservation. Documenting unidentified calls, establishing inventories of biological sounds, and developing generalizable machine learning models contained within this Research Topic provide insights into the emerging tools and baselines that advance our ability to monitor aquatic ecosystems during an unprecedented global biodiversity crisis. As we move forward, the continued collaboration and open sharing of data through repositories and global libraries will be the cornerstone of our ability to listen to, and ultimately protect, the symphony of life beneath the waves.
Iorio et al. (Tue,) studied this question.