Birds2016
Japanese great tits showed different receiver behaviors under component calls, natural versus reversed order, and single- versus two-source playback; a bounded novel cross-species sequence also elicited stronger responses than its reversal, while a matched ABC–zi control was weak. In the 2022 deposited binary table, the natural single-source 1A-R condition had an approach event in 13/16 flock trials and wing-flicking in 15/16; the other three treatments had approach events in 2–3/16 and wing-flicking in 1–4/16. These aggregate counts reproduce the descriptive pattern only. They do not establish a unique mechanism: one published critique agrees that receiver behavior is sensitive to order and source but disputes that this demonstrates Core-Merge or hierarchical syntax.
Primary evidence and alternative theory reviewed
Marine mammals2024
Analysis of sperm whale codas describes contextual and combinatorial acoustic structure.
Initial source note
Marine mammals2026
A reported sperm whale birth connects changing vocal styles with an observed social event.
Initial source note
Machine learning · bioacoustics2025–2026
NatureLM-audio combines a BEATs audio encoder with Llama 3.1 8B Instruct and responds to audio-plus-text prompts with outputs such as species labels, detections, call types, life-stage classes and captions. Its ICLR 2025 paper reports strong results on several BEANS-Zero tasks and evaluates held-out taxa; the paper cautions that BirdNET and Perch comparison datasets overlap their training data, so those are topline rather than fully fair comparisons. A NeurIPS 2025 workshop follow-up preprint exposes prompt-format brittleness: asking for common and scientific names together reduced accuracy to 6–12% on two benchmark components compared with prompts asking for one name. For those combined-name prompts, the authors report that LoRA interpolation at about α=0.7 raises accuracy from 6% to 45% on Watkins and 12% to 63% on CBI. Separately, on one 425-sample, 40-class closed-set unseen-family task, reported F1 rose from 0.09 to 0.28 (the authors describe this as a >200% relative improvement). These are bounded classification and instruction-following results, not evidence that the model discovers what an animal means or can hold a conversation with it. The current training-data viewer lists 26.4 million rows; the paper notes that prompt counts can derive from the same audio, so row counts are not unique recordings. The official BEANS-Zero card reports 91,965 examples in a single test split and says the composite is not intended for commercial use; it documents a license field per sample. The live Hugging Face tree lists 50.5 GB, while the companion GitHub README says about 100 GB for a complete non-streaming download. Its tasks cover classification, detection, call-type/lifestage and captioning, not receiver comprehension or communication meaning.
Targeted primary-paper and official model/data/code metadata review; no independent reproduction
Machine learning · bioacoustics2026
BEANS-Next evaluates 45 tasks with 37,718 examples across acoustic perception, biological-category recognition, scene understanding and in-context learning. ROOTS is a separate training resource reported to contain 43,922,135 audio-language conversations. The official project overview describes approximately 44 million pairs from 6.5 million distinct audio clips; the current Hugging Face card says its audio resolver index contains 4,745,972 distinct references. The count definitions have not been reconciled and neither is a count of independent natural animal interactions. The authors describe BEANS-Next evaluation examples as real recordings paired with human- or dataset-provided annotations; text templates and LLMs format the supported annotations into prompts rather than creating new benchmark labels. They state the benchmark audio and derivatives are disjoint from ROOTS and use held-out species where applicable. ROOTS combines curated recordings and structured/free-text metadata with audio-derived information and synthetic audio scenes. The paper reports capability gains after tier-specific fine-tuning, while its overview still identifies weaknesses in open-ended description, counting and transfer to real-world recordings.
Targeted primary-paper and official metadata review; no data acquisition, model run or independent reproduction
Machine learning2026
BirdNET supplies bird-sound identification tools and a human review workflow for detections.
Documentation checked
Machine learning2025
Perch v2 provides acoustic embeddings for downstream classifiers. The official CPU version 1 card declares Apache-2.0 and says its weights match the GPU perch_v2 version 2. Pinned Hoplite configuration uses five-second, mono 32 kHz windows and a 1,536-dimensional embedding.
Documentation checked
Marine mammals2025
DolphinGemma models patterns in dolphin sounds. The related CHAT work uses learned shared signals in human-dolphin interactions.
Initial source note
Companion animals2024
The study reports 62.18% accuracy for Wav2Vec2 pretrained on 960 hours of human LibriSpeech, versus a 56.37% majority baseline, on a selected four-class bark/context task. The result is prediction of association with induced study situations, not a test of what dogs understand or mean.
Full paper methods reviewed
Birds2024
In a six-month, nine-parrot home study, Table 2 lists 65 parrot-triggered live calls and 40 playback calls out of 108 possible in each phase. Recalculation from the displayed counts gives live-minus-playback differences of +50, +17, +75, +17, +8, -8, -8, +33 and +25 percentage points for P1–P9. The printed row is reversed for seven birds and appears to swap the P3/P5 entries. The authors report a live-phase advantage, but the published duration totals do not reconcile across the text and Table 3; the duration magnitude remains unresolved.
Methods reviewed
Companion animals2024
Research-team reporting describes appropriate responses by trained dogs to selected soundboard words.
Initial source note
Plants & fungi2023
Khait et al. report airborne ultrasonic emissions from stressed tomato and tobacco in an acoustic box and greenhouse, and classify plant condition from recordings. The system used paired CM16 ultrasound microphones per plant at 500 kHz, with a 20 kHz high-pass filter; an event counted as plant-associated only when both microphones aimed at that plant recorded it. They report leave-one-plant-out evaluation: a 1-hour greenhouse dry-versus-irrigated classification reached 84% balanced accuracy, and a two-range soil-moisture classification in a nine-day dry-down of 23 tomato plants reached 81%. These are emission and physiological-state results, not decoded messages or tested receiver responses.
Full paper, critical context and component-rights metadata reviewed
Plants & fungi2026
In controlled choices, female Egyptian cotton leafworm moths (Spodoptera littoralis) changed where they laid egg clusters when playback of drought-stressed tomato-plant clicks was present. The preference depended on the other cues: playback attracted oviposition in a plant-free arena, while moths preferred the silent side when healthy tomato plants were offered on both sides. Deafening removed the plant-free playback preference; male-moth click controls did not produce a detected preference. A separate long-arena experiment used a lower 30-click/min rate and found egg clusters concentrated near the playback and feeder rather than the silent edge. A checksum-matched aggregate audit of the single Mendeley data.xlsx file also matched the five main Table 1 tuples for repetitions, paired arena/date observations and egg-cluster totals; this is a source-table consistency check, not a statistical-model reproduction.
Full-text methods, results and public review record reviewed; one checksum-pinned source table checked in aggregate
Plants & fungi2025
Measurement work and critiques highlight difficulties in separating fungal electrical activity from environmental artifacts.
Initial source note
Fusarium oxysporum (filamentous fungus)2025
Buffi et al. tested a printed fungal-potential card on one Fusarium oxysporum strain (NEU 195). Table 1 reports a significant pre-induction power-spectral-density difference of about 1,604.752% between 24 fungal electrode-pair records and 8 non-inoculated control records (p<0.001). The study supports a specialized laboratory measurement of fungal-associated electrical activity; it does not decode a fungal language.
Full primary article and dataset metadata review; supplement and files unreviewed
Research tools2026
Arbimon, BORIS and BirdNET review tools already cover parts of acoustic analysis, behavioral annotation and detection review.
Documentation checked
Machine learning2026
Public benchmarks cover different tasks: the BEANS dog labels identify individuals, while BEANS-Next includes acoustic-description tasks.
Documentation checked
Birds2023
A study of 229 monk parakeets tests caller identity across call types; it does not predict behavior or decode meaning.
Targeted study and dataset-fit review
Cross-species2022–2026
Self-supervised audio encoders learn from unlabeled recordings. AVEX provides a common interface for extracting features and fitting small supervised classifiers.
Abstract & documentation reviewed
Birds2023
BIRB tests retrieval from passive soundscapes using examples from focal bird recordings, including rare and held-out species.
Selected sections reviewed
Cross-species2021–2026
Deep Audio Segmenter connects manual labels, model training, automatic proposals and human correction in an existing audio workflow.
Documentation reviewed
Bats2016
This study paired bat social calls with synchronized video to label caller, recipient, circumstances and subsequent behavior.
Selected methods reviewed
Mammals2024–2026
The peer-reviewed study presents animal2vec as a rare-event bioacoustic detector/classifier and MeerKAT as 1,068 hours of 8 kHz meerkat audio, including 184 hours with complete event labels across 66,398 ten-second samples. The paper reports eight vocalization classes and three non-call event classes, overlapping labels, millisecond event boundaries and inter-annotator timing reliability just over 10 ms. It is a valuable example of continuous audio, quiet/background intervals and event-based precision–recall evaluation.
Targeted peer-reviewed methods, results and rights review
Elephants2024–2026
The 2024 study combines acoustic classification with receiver-response experiments to investigate individually directed calls.
Selected methods & critique notice reviewed
Across animals2022–2024
Pose tracking and supervised behavior tools offer building blocks for connecting a call to visible movement.
Initial source note
Plants & fungi2023
Laboratory Arabidopsis imaging connects particular airborne leaf volatiles with calcium responses and defense-related gene expression.
Initial source note
Plants & fungi2026
Adamatzky (2026) reports slow, ordered electrical events across eight differential channels placed along one Pleurotus ostreatus-colonised substrate, recorded for about fifteen days. The paper interprets lead–lag structure as propagation consistent with ionic waves.
Full-text methods and data-provenance review; no reproduction
Plants & fungi2021–2025
Adamatzky's peer-reviewed 2022 paper reports extracellular voltage recordings from four fungal species. It sampled one value per second from differential electrode pairs placed 1–2 cm apart, recorded up to eight pairs at once, and reports about 1.5 days for Schizophyllum commune and about five days for each other species. The analysis used species-specific thresholds to identify spikes, then grouped them into proposed “words” using one or two times each species' mean inter-spike interval. The resulting spike-count distribution was compared with human-language word lengths. This is a published pattern-analysis hypothesis, not a decoded fungal vocabulary or conversation.
Full preprint methods and 2025 review; critique abstract/selected sections; dataset metadata only
Scientific method2006–2017
The 2006 pilot reports that 100 blind judges gave higher beauty ratings to 24 selected treated-condition images than to 16 control images (means 2.87 vs 1.88; one-tailed p=.001). The images came from just four bottles (two per condition), and only drops with a discernible crystal were photographed. The same investigator group’s 2008 triple-blind follow-up used six bottles (two per condition) and 300 drop photographs. Its authors reported a weak three-condition beauty effect (p=.03) and a treated-versus-proximal-control comparison at p=.05, one-tailed; the beauty effect was not significant after adjustment for image interest (p=.20), and the three-condition objective-contrast result was not significant (p=.25). These are reports by the study authors, not independent confirmation.
Targeted full-text methods review; not a systematic review
Parrots2022
The authors received 898 voluntary survey responses and retained a cleaned sample of 877 companion parrots across 33 genera and 73 species after quality checks and excluding birds not reported to mimic human sounds. Respondents said sampled parrots sometimes (36%) or frequently (53%) used mimicked vocalizations in appropriate human contexts without prompting. These are caregiver reports, not independently observed events or receiver tests.
Full main-text review and metadata-only dataset/privacy assessment
Parrots2026
The survey reports on 1,202 companion parrots across 89 species; phrase-use information was supplied for 884 birds, with names reported for 413. Multiple observers identified 131 reported examples of appropriate name use across 88 birds and 30 species, and 69 individualized examples across 42 birds and 19 species.
Targeted primary-source review
Parrots2026
PLOS describes S2 as a spreadsheet of parrot name phrases with coding and de-identified supporting information. The paper says phrase-use context was based on survey-takers' accounts of when or how a phrase was used, not inferred from the phrase's words; observers coded and reconciled the reports. The NLM catalog reports CC BY 4.0 for the linked Figshare item.
Metadata-only source and rights review
Parrots2023
The full article used 16 adult budgerigars, with eight in the contact-call cohort and five in the neural-inactivation experiment. Playback of colony-noise segments elicited 180–1,612 contact calls per bird per day in the eight-bird cohort. A VAE representation and SVM identified callers at over 90% accuracy in a balanced random 80/20 call-level split, repeated 50 times; accuracy remained about 90% for two shared call types. A classifier trained at week 3 and tested at week 15 fell from about 90% to 40%, consistent with call change over time. In the separate five-bird experiment, birds still responded to individual calls during AFP inactivation, although response rate fell in three birds and latency effects varied. Inactivation altered call acoustics and reduced individual discriminability. These are bounded acoustic identity and production findings, not decoded meaning.
Full article methods and results reviewed; dataset remains a one-file exploratory sample
Cats2020
CatMeows documents 440 sounds from 21 cats with context, cat, owner and session identifiers.
Selected evidence reviewed
Farm animals2022
The paper analyzes calls across production contexts. The open Soundwel subset contains 6,888 calls in 17 categories.
Selected evidence reviewed
Amphibians2023
AnuraSet describes 27 hours of expert annotations across 42 Neotropical frog species. The paper reports a 70/30 iterative multilabel split grouped by one-minute recording and a 37.8% overall macro-F1 at threshold 0.5; the rare-species group scored 15.7%. This is a species-identification task, not a call-meaning benchmark.
Selected evidence reviewed
Insects1973
Von Frisch’s lecture connects waggle-dance duration and orientation with food distance and direction. It also recounts a 1949 hive-transfer experiment probing orientation with unfamiliar surroundings.
Selected primary passages reviewed
Fish2019
A field experiment compared reef-sound playback with dummy-speaker and no-speaker controls. Juvenile fish recruitment and retention increased on acoustically enriched experimental reefs.
Selected primary passages reviewed
Birds2016
Researchers found that a specialized honey-hunting call used by people in Mozambique increased cooperative guiding by wild honeyguides compared with other human or animal sounds.
Selected primary passages reviewed
Mammals1980
The 1980 publisher summaries report three Amboseli groups observed over 14 months and predator-absent alarm-call playbacks; the longer companion abstract says call length, amplitude and caller age/sex had little effect on response quality, while context was not a systematic determinant in that experiment. The 1980 Science abstract reports different receiver behaviors to predator-call playbacks. Fischer (2020) separates comparatively constrained call production from learned receiver responses. In a later South African field study, eight adult female subjects from one group completed 16 within-subject playback trials using two alarm-bark exemplars from one male. Subjects looked toward the speaker for less time and showed lower vigilance during an ongoing intergroup encounter than when no encounter was underway; the authors reported no predator-specific escape responses. This is evidence that surrounding context can change measured responses to a call in this bounded setup.
Selected evidence reviewed
Marine mammals1985
Across 163 songs sampled near Bermuda, songs within a year were similar while their content changed across years. The series covers 13 sampled years between 1957 and 1975.
Selected evidence reviewed
Birds2011
Researchers observed ravens showing or offering non-food objects to other ravens, followed by attention to the object and social interaction.
Selected evidence reviewed
Mammals2013
In two captive wolf packs, howling after separation was associated with social relationships. The measured cortisol difference did not explain howling frequency within the test condition.
Selected evidence reviewed
Fish2006
Groupers made visual movements before coordinated hunting with moray eels. The study links signaling to hunting context and reports benefits from complementary hunting behavior.
Selected evidence reviewed
Marine mammals2013
Dolphins responded vocally to copies of their own signature whistles more than to the control whistles described in the study, supporting individual-directed use of learned signals.
Selected evidence reviewed
Parrots2024
In the study's specific 10.1-inch tablet task, 17 trained pet parrots completed 4,720 selected target touches (8,995 counting follow-on taps). Hit rate increased with target width from 19.29% at 40 dp to 47.08% at 130 dp; the study found no significant effect of target distance and no detected difference in touch-point precision by target width. The authors' 300 ms rule classified 45.73% of taps as multi-taps. This is touchscreen-ergonomics evidence, not a communication test.
Full published article text reviewed
Parrots2023
In a 48-bird study, older adults showed reduced vocal diversity over the experiment and older-age flocks had weaker/less dense affiliative ties; measured vocal plasticity and convergence were similar across age classes. The deposit pairs derived call features and social observations at flock/date/block scale, not at synchronized call-event scale.
Primary methods and dataset-fit review; files remain metadata-only
Mammals2025
MammAlps provides a real multimodal wildlife behavior benchmark: animal-centered video clips have matched audio and scene maps, with visible behavior labels assigned by two annotators. It evaluates observable actions and activities, not vocal meaning or animal translation.
Targeted paper, release and benchmark-fit review
Parrots2022
The study reports vocal babbling and repertoire development in free-ranging green-rumped parrotlet nestlings of both sexes. The paper describes a corticosterone-related repertoire difference and says the babbling repertoire was not yet integrated into functional contexts; this is developmental biology, not evidence of call translation.
Targeted paper and dataset-lead review
Mammals2026
Prompting-MammAlps tests whether natural-language queries can retrieve camera-trap videos whose structured, visible-behavior annotations satisfy each query. The 135 ecology-informed queries include multi-attribute, multi-animal and temporal relations. This is a text interface to a video-label schema, not translation of animal signals or a two-way communication system.
Targeted paper, benchmark, release and rights review
Egyptian fruit bat (Rousettus aegyptiacus)2017–2023
The 2017 archive describes 293,238 Egyptian fruit-bat WAV files recorded at 250 kHz and 16-bit, plus 91,080 labels derived from synchronized video, including caller, addressee, behavior and social context. The data descriptor reports a post-hoc recheck of 435 annotations, with a 4% estimated context-label error rate (95% CI 2.2–5.8%). The reviewed metadata PDF says annotation FileIDs map to FileInfo.csv; its tables describe recording-date ranges, treatment/channel settings and bats that recur across treatments, and define known, unknown and ambiguous caller/receiver codes. This makes a file/animal grouping audit plausible, but not yet verified against FileInfo rows.
Targeted primary literature, dataset and rights review
Yellow-naped amazon (Amazona auropalliata)2026
The study examined warble duets from 13 mated pairs of wild yellow-naped amazons in north-west Costa Rica during 2006–2008. The authors report creating 1,136 duet files; 52 high-quality warble duets from 12 pairs were used in the syntax analysis. They report recurring call types, flexible ordering, 19 positive and four negative call-type associations, while most complete duet sequences were unique. This is evidence of measurable sequence structure in this study, not a translated conversation.
Targeted primary article, dataset metadata and reuse-rights review
Orange-fronted conure (Eupsittula canicularis; the 2012 paper uses Aratinga canicularis)2012
Balsby, Momberg and Dabelsteen tested whether a call resembling one of two paired birds would selectively affect that bird. They studied 36 temporarily captive wild conures from 27 flocks. After exclusions, 44 of 72 trial periods were analyzed. Targeted birds responded with higher call rates and shorter latency overall, while some non-target responses were also recorded. The authors interpret this as evidence that vocal imitation can address a specific individual in this species and setting.
Full-text methods and results review
Egyptian fruit bat (Rousettus aegyptiacus)2024
Triantafyllopoulos et al. trained models to predict 11 video-coded interaction contexts from Egyptian fruit-bat calls. Their best reported result was 33.3% unweighted average recall; this predicts observed context, not message meaning or receiver comprehension. A fresh checksum-verified audit of the paper-described subset matches 35,074 annotations. Strict source-file-plus-participant graphs remain highly connected (largest 95.56%), but a role-aware caller diagnostic excluding 122 mixed-caller source files retains 34,815 annotations (99.26%) and leaves every context represented by 8–41 callers. This makes a caller-held-out test plausible for review, not validated: the paper already reports subject-independent folds, exact membership is unverified, caller identities can appear as addressees in development, and session/chamber/treatment factors remain unjoined. No split or model was created.
Paper, dataset, split-feasibility and reviewer-independence review
Rhesus macaque, bottlenose dolphin and Egyptian fruit bat2021
BioCPPNet is a supervised source-separation pipeline, an audio-cleanup step rather than a communication decoder. The study constructs mixtures by additively combining known individual vocalizations, then trains a lightweight U-Net to estimate separated waveforms. It evaluates macaque, dolphin and Egyptian-fruit-bat material. For bats, the paper reports 31,399 calls from 15 highly represented individuals, with 12 used in a closed-speaker regime and three held for an open-speaker evaluation. It reports bat validation SI-SDR 9.7 and downstream caller-identity accuracy 57.5% for the closed-speaker setting; the open-speaker SI-SDR is reported as 10.0. Those results concern reconstruction and identity on generated mixtures, not behavior, context meaning, receiver comprehension or conversation.
Primary paper method/results review and pinned implementation-repository review
Multiple taxa; the review discusses spectacled parrotlets, common marmosets, African elephants and bottlenose dolphins2026
Jaakkola, King and Bruck propose three minimum criteria for identifying animal signals as names: individual reference, symbolic representation and shared meaning. Their abstract says the review considers evidence about name-like calls in spectacled parrotlets, common marmosets, African elephants and bottlenose dolphins; publisher highlights say much of the presented evidence is equivocal. This is a conceptual review, not a new experiment or an ML result.
Abstract and publisher-highlights review; full text was not accessible in this review
Birds · Dupont's lark (Chersophilus duponti)2026
The article describes standardized recordings from 20 Spanish populations and reports 4,297 annotated songs from 191 singing males, representing 401 song types, plus 795 calls from 97 calling males, representing 80 call types. Experts categorized acoustic song and call types and provide event timing, frequency bounds and individual IDs. One ornithologist performed the primary annotation; uncertain classifications and types reported as shared among populations received additional expert review. Zenodo resolves the article's cited concept DOI 10.5281/zenodo.17264916 to current version record 10.5281/zenodo.17264917; the API identifies 17264916 as conceptdoi, 17264917 as the version doi, and declares record-level CC BY 4.0. Its eight-file metadata inventory includes provider-supplied sizes and MD5 values. The separate Digital.CSIC distribution also identifies version DOI 17264917 and declares CC BY 4.0. The CSIC archive bytes and their relationship to the Zenodo file set remain unverified.
Full publisher article, Zenodo concept-DOI redirect and version metadata, public-sector catalog, full CSIC README and archive HEAD metadata reviewed; archive and code bytes not inspected.
Plants & fungi2023–2025
A common mycorrhizal network describes fungal hyphae shared by roots of two or more plants; such connections have been documented in particular systems, while their prevalence and functions remain debated. Whether material moves through them, whether that movement changes a receiver, and whether information is communicated are separate questions. Karst, Jones and Hoeksema (2023) warn that popular and scientific claims about widespread forest networks and beneficial “mother tree” transfers can outrun the evidence and report positive citation bias. Henriksson et al. (2023) argue that isotope studies do not yet establish significant net carbon transfer through these networks that benefits recipients, and discuss alternative pathways. Simard, Ryan and Perry (2025) respond that multiple methods support network connections and context-dependent transfer in forests. This published disagreement is a reason to state the evidence level precisely, not to declare either that networks never occur or that trees have a demonstrated shared language.
Targeted publisher-source review; not a systematic review
Great green macaw (Ara ambiguus)2026
Allison, Jimenez and Wright compare great green macaw adult contact calls across wild, captive and reintroduced Costa Rican populations and describe nestling call development. The publisher abstract reports minimal adult call variation among the three population categories and variation in nestling calls during mid-development. The authors discuss morphology, call overproduction, selective attrition and vocal convergence as possible explanations. This concerns acoustic development and variation, not a demonstrated call meaning.
Targeted publisher abstract and dataset-metadata review; full methods not reviewed
Cross-species: rock hyrax, zebra finch, little owl and infant marmoset2026
Arias-Vergara et al. compare 15 frozen speech, speaker-verification, animal-bioacoustic and general-audio encoders on caller identification in four species and call-type classification in three. The strongest speech-pretrained representation outperforms animal-pretrained representations in three of seven task–dataset comparisons; the other four differences are not statistically significant. Controlled comparisons find no systematic advantage from multilingual coverage or animal-domain pretraining, and useful encoder layers vary with architecture. This shows that some acoustic caller and call-type information transfers across domains; the study does not predict independently observed behavior or test message meaning.
Full arXiv v1 HTML and version/license metadata reviewed; associated datasets, code and model assets not audited
Monk parakeet and yellow-naped amazon2023
The authors report stronger acoustic convergence at the individual than site scale in native-range Uruguay and introduced U.S. monk parakeet populations. Their yellow-naped amazon comparator showed stronger convergence at the regional-dialect scale. This is evidence about acoustic identity structure and a proposed identity-signaling framework, not decoded call meaning or demonstrated receiver comprehension.
Targeted primary article, dataset metadata, and code-provenance review; no file acquisition or reproduction
Cross-species bioacoustics2025
Whombat is a browser-based workbench for managing bioacoustic projects, reviewing recordings and human annotations, and visualizing or exchanging machine-learning predictions. The paper describes bat and avian applications. Its guide imports an AOEF dataset with an audio directory first, then its annotation project; evaluation sets and model runs are separate, and predictions are distinct from human annotations. Talk2Nature now offers an optional experimental one-way AOEF 1.1 package. The exact export imported into Whombat 0.9.0 with one recording, intervals [1, 2] and [4, 5] seconds, four visible namespaced project tags, and two notes. A missing AOEF project_tags list and a macOS canonical-path issue were identified and fixed/documented during this synthetic compatibility test.
Article and guides reviewed; synthetic AOEF 1.1 export loaded with soundevent 2.10.0 and imported through Whombat 0.9.0 desktop UI; specialist review pending
Cross-species bioacoustics2026
The project describes a shared Python interface for more than 35 bioacoustic datasets across several animal groups, with streaming, configurable transforms and per-dataset metadata. Its README explicitly leaves each dataset under its original creator’s terms.
Official repository overview and exact PyPI release metadata reviewed; integration and individual dataset rights not audited
Monk parakeet; cross-species bioacoustics method2022
ANIMAL-SPOT is a staged, species-specific acoustic detection/classification framework, not one zero-shot animal-language model. Its paper reports recording-level partitioning (samples from a recording stay within one split) and a default 70/15/15 train/validation/test split. For target/noise detection, it reports mean test-set accuracy of 97.9% (range 94.5–99.8%) and, separately, an average AUC of 95.9% (range 91.7–99.1%) across 33 additional unseen animal-specific recordings. In a separate monk-parakeet task, the paper reports 92.7% call-type accuracy on human-labeled test samples and 88.4% on detector-selected segments. These are upstream-reported results, not Talk2Nature measurements; acoustic class prediction does not establish behavior, receiver comprehension or meaning.
Full main-article text and split methods reviewed; current repository access terms and API field semantics rechecked but not accepted; supplement, archive contents and specialist review remain open
Cross-taxon bioacoustics2026
The directory is a useful cross-taxon discovery lead and a reminder to record dataset version, license, annotation level, recording type, geography and other metadata. The project says missing information is being added over time, so its entries should be treated as leads that require verification at the dataset's original source.
Pinned repository README, root inventory and API metadata reviewed; individual dataset records and underlying rights not audited
Ecoacoustics · machine learning2026
CoarseSoundNet targets broad, potentially co-occurring soundscape layers: anthropophony (human-made sounds), biophony (animal sounds), geophony (natural abiotic sounds) and an explicitly handled silence condition. The manuscript reports strong EDANSA-2019 test performance but a marked drop on the different BEsound recording domain, especially for human and geophysical sounds. Its best BEsound macro-F1 of 0.799 applies class-specific thresholds and window-count settings selected on BEsound itself; it should be treated as a tuned evaluation result rather than an untouched holdout estimate. The ecological case study uses model predictions to pre-filter audio before calculating conventional acoustic indices. This can help characterize recording context, but it does not say which species vocalized or what a call means. The paper uses public EDANSA together with BEsound and other collections that are not all directly accessible. Code, weights, article and data have distinct licenses: the pinned training repository declares MIT, the Hub model declares CC BY-NC-SA 4.0, the arXiv paper declares CC BY-NC-SA 4.0, and Zenodo's API declares CC BY 4.0 for the EDANSA record while the rendered license field is blank.
Full preprint review and official code/model/dataset metadata inspection; no independent reproduction
Sperm whale · cetacean bioacoustics2025
WhAM fine-tunes the music-pretrained VampNet masked acoustic token model in two stages: domain adaptation on animal-audio collections, then sperm-whale-specific training on DSWP and CETI. The paper describes 2,507 DSWP and 7,653 CETI annotated codas for this work. It evaluates generated audio with Fréchet Audio Distance and a five-expert listening study, and tests learned embeddings on coda detection, rhythm, social-unit and vowel-type classification. The paper explicitly defines “translation” as converting prompts, including noise or other animal sounds, into sperm-whale coda-like acoustic style; it says semantic translation remains a separate, more ambitious goal. On the reported classification table, WhAM exceeds simple random/majority baselines but the reported specialist or general bioacoustic embeddings perform better across these tasks. For example, social-unit accuracy is 70.5% for WhAM versus 93.2% for BirdNET. The paper describes social-unit classification within data from a single vocal clan as closer to speaker identification than dialect classification.
Full main-text review and targeted methods/artifact metadata review; no independent reproduction
Cross-taxon biodiversity data2026
Darwin Core provides a biodiversity-data vocabulary and conceptual model; its 2026 Data Package guide specifies a descriptor, tabular resources and schemas whose field descriptors must come from the official per-table schemas. Audubon Core describes biodiversity-related multimedia resources and offers terms for aspects such as rights and access. Issue #759 is closed, and its visible discussion says guide section 3.4 already states the field-descriptor rule; the official review index was stale. The separate deployment gap remains: the versioned profile/schema URLs returned 404 here, while the repository's visible snapshot is unversioned and marked pre-review. These standards may support future metadata exchange, but the present Station session is not yet a protocol-defined sampling event.
Targeted review of official standards documentation, the TDWG schema-source commit and issue #759; no Talk2Nature conformance prototype or independent standards review
Birds2025
Great tits approached the loudspeaker more often during call playbacks than during control playbacks, but approach did not significantly differ among call treatments. Wing-flicking was more likely after the complete two-part call sequence (labelled FME–D by the authors) than after D-only calls or controls for both great-tit and Carolina-wren stimuli. The study reports no significant difference between the two species’ sequence treatments (P=0.12; wide confidence interval), so it does not establish their equivalence.
Publisher-indexed methods/results and institutional abstract reviewed; no article file, data, or independent specialist review
Plants & fungi2026
In split-plate laboratory experiments with Arabidopsis thaliana, Waschk et al. report that three tested endophytes formed common endophyte networks and moved some isotope-labelled resources from donor-side soil into receiver plant tissues. The paper reports 55% higher amino-acid-derived 15N in receiver plants for Fusarium temperatum and a smaller 13C enrichment for Trichoderma viride (+2.83% over control); it also reports 38% greater plant biomass for F. temperatum in this setup. These are bounded physiological and isotope outcomes, not a decoded signal or a demonstrated plant conversation.
Full primary article and repository metadata reviewed; no artifact reproduction or independent specialist review
Lilac-crowned Amazon (Amazona finschi)2016
Montes-Medina, Salinas-Melgoza and Renton analyzed 75 hours of non-continuous field recordings from 61 free-living Lilac-crowned Amazons. They report 8,622 notes classified into 152 note-types; after excluding single-occurrence types, 8,571 notes formed 101 note-types emitted at least twice. The researchers associated calls with nine behavioral contexts, and common note-types occurred in several contexts with different proportions. Context was associated with acoustic features and note-type frequency. Separately, a nested permuted discriminant analysis distinguished 28 note-types while accounting for repeated notes from individuals (46.9% observed classification versus 9.6% expected under the paper's permutation null, P=0.001). That is note-type discrimination, not prediction of behavior from sound and not evidence that calls have human-language meanings.
Full-text primary-source review
Multiple taxa: common nightingale and bats2021–2024
Two different studies show that smartphones can support bounded wildlife-acoustic tasks, but they do not create a uniform sensor. Jäckel et al. compared 5,679 Berlin citizen-science nightingale recordings with expert recordings: 53% of citizen recordings were categorized as complete songs versus 100% of expert recordings, and signal-to-noise ratios were lower for citizen recordings. In a small controlled playback comparison, estimated frequency limits varied by device. The authors nevertheless considered the citizen recordings useful for questions based on relative song-type occurrence, while cautioning about exact acoustic measurements. Gili et al. tested built-in microphones for selected 8–24 kHz bat calls. Their field/pilot work covered 35 phone models and nine bat species, but found device-model variation, occasional aliasing and loss of calls near the upper detection limit; the authors recommend preliminary detectability checks and expert verification of at least some identifications.
Selected full-text methods, results, discussion, data-availability and rights review
Birds2026
PteroSet v5 is a useful time-frequency bird-event detection comparator: its released record describes 563 time-lapse WAVs totaling 73.62 hours, 15,372 expert annotations and 6,702 species-level determinations across 168 species. It offers a binary bird-presence baseline. That target is whether a window overlaps an annotated bird vocalization, not what a bird is communicating. The dataset supplies no independently observed behavior, interaction, recipient response or communicative-meaning labels.
Full preprint methods and evaluation; versioned dataset manifest and header-only metadata schema; pinned code review
Cross-species bioacoustics2023
OpenSoundscape is a Python research toolkit for manipulating audio and spectrograms, working with Raven time-frequency annotations, preparing labels, training or applying acoustic classifiers, detecting periodic calls, and localizing sounds from synchronized recorders. Its 2023 paper describes version 0.9.1; the current official release page lists v0.13.2. This is useful infrastructure for acoustic observation and researcher workflows, not a model that translates animal communication. The linked Dryad dataset is a distinct resource: its current landing page reports 385 minutes of bird soundscape recordings across four days, 48 species and 16,052 strong annotations. These labels support bounded detection or classification tasks; they do not establish call meaning or receiver understanding.
Selected paper sections, versioned repository/license, annotation tutorial, dataset record and Dryad policy reviewed; no code or data executed
Spotted hyenas2026
HyenaSET is a useful recent sound-event methods comparator: its Edmond v1 metadata describes 590,480 ten-second clips, about 1,640 hours from 19 spotted hyenas, with 87,394 expert-reviewed clips (about 243 hours). The annotations mark vocalization intervals and types, and distinguish calls from the collar-wearing focal hyena from nearby non-focal callers. The preprint provides sample-stratified splits and a four-individual held-out split, then reports an animal2vec baseline. These targets concern event detection, call categories and source attribution; they are not independently observed receiver responses or evidence of message meaning.
Targeted preprint methods and results; official dataset version, file and terms metadata
Cattle2026
CowRumIA is a 2026 undergraduate project pairing a Flutter Android app with YAMNet embeddings and a dense classifier for four cattle vocalization categories. The Zenodo v1 record describes 624 labeled WAVs; the README reports 194 stressed beef, 39 calm beef, 350 stressed dairy and 41 calm dairy samples and advertises 85.6% accuracy on 125 test audios. In the pinned training script, a random stratified 80/20 utterance split creates X_test, then the same X_test and y_test are supplied to training as validation data while callbacks select epochs and adjust the learning rate; predictions are later evaluated on that same set. The reported score therefore is not an untouched final-test estimate. The size or direction of any resulting optimism cannot be determined from this review. This is a useful mobile-implementation comparator and a strong evaluation-design lesson, not evidence of animal communication or a validated welfare diagnostic.
Pinned GitHub README, LICENSE and training-script static review; official Zenodo v1 metadata review
Birds2016
Five individually housed budgerigars had been trained to call after a visual LED cue, with food reinforcement. In a call-plus-flock condition, calls were louder than in silent, call-only, flock-only, and pure-tone conditions. Among the three birds that produced non-primary calls, these calls were more likely in the call-plus-flock and call-only conditions than in silent, flock-only, and tone conditions. The result shows acoustic context can modulate a trained vocal response in this laboratory setup; it does not decode a message.
Selected methods and results reviewed
Insects2017
Wario et al. built a pipeline to detect honeybee waggle runs from a one-frame observation hive, estimate run orientation and duration, and map dances back to the field. The experiment used a small colony of 2,000 Apis mellifera carnica and one modified high-speed camera on each side of the comb, with infrared illumination. The paper specifies about 100 frames per second and at least 1.5 pixels per millimeter. Its temporal CNN filtered candidate image sequences; 8,239 manually labeled detections from two days were used, with 20% reserved for validation. The authors report 90.07% validation accuracy, corresponding to 89.8% recall at 95% precision. Orientation was compared with eight human tracers on 200 waggle runs, with mean error −2.92° and SD 7.37°. In a five-hour illustration, 571 detected dances from bees trained to a feeder 342 m southwest of the hive clustered around the feeder direction. This decodes a predefined, previously studied signal against a known physical outcome; the algorithm does not discover what arbitrary animal movements mean.
Primary paper, pinned artifact metadata and separate system comparator reviewed
Dogs2025
Barkopedia was an IJCAI 2025 challenge with eight dog-audio tasks: age, sex, breed, individual recognition, activity/environment, emotion, vocal separation and vocal detection. Its README describes the corpus as real-world clips sourced from YouTube and Reddit. The reviewed Activity and Environment card lists 12,480 training clips and 3,120 test clips, with eight activity categories and six environment categories. It says labels were initially generated through video-assisted inference with Janus-Pro-7B and later manually verified. This is a useful context-label classification comparator, but it does not show that an audio prediction represents an animal's intended message, subjective state or a validated welfare outcome. The pinned Hugging Face tree includes train, validation and test directories, although the card still says the test set will be released later. Its Dataset Viewer reports incompatible audiofolder/CSV formats across splits. The Barkopedia GitHub repository is the MIT-licensed challenge site/documentation; its inspected root tree did not include a training or evaluation implementation.
Official challenge and one task-specific dataset card reviewed; Task 6 card history checked only; metadata-only
Yellow-naped amazon (Amazona auropalliata)2024
The paper's abstract reports that learned yellow-naped Amazon contact-call dialects shifted between 2005 and 2016: more sites used multiple call types, formerly distinct northern and southern regions became more acoustically similar, and new variants and greater acoustic variation appeared. The authors place these observations alongside a period of population decline. The version-6 Dryad deposit documents measurements and metadata for surveys in 1994, 2005 and 2016, but not the underlying audio.
Abstract-level paper review plus Dryad dataset/API and pinned-code metadata review
Domestic dog (Canis familiaris), Siberian Husky and Shiba Inu2025
The paper predicts two three-class labels from audio: arousal (low/medium/high) and valence (negative/neutral/positive). Annotators first watched the associated video with surrounding context and used visible behavior and situation to assign the labels; samples were retained only when at least two of three raters agreed on both dimensions. These are human video-derived behavior/affect annotations, not an animal-reported state, an independently validated physiological measure, a receiver response, or evidence that a bark means a particular thing. The article reports dog-separated tests with best accuracies around 50–57% across breed and target combinations. It separately reports higher 10-fold cross-validation results, but the reviewed methods do not specify whether those folds were grouped by dog. The Hugging Face path formerly used for Barkopedia Task 6 was rewritten at the pinned revision as EmotionalCanines and its license metadata changed; no continuity of the underlying files was verified.
Targeted full-text methods/results review; version-pinned dataset metadata review; no independent reproduction
Birds, mammals, amphibians and insects2025
El Silencio is a useful multi-taxon sound-event detection candidate: the paper describes 770 one-minute soundscapes (12.8 hours) and about 15,000 event boxes across birds, mammals, amphibians and insects. The manual protocol labels acoustic group, call/sonotype and sound form; it groups same-species signals separated by no more than one second. These acoustic/taxonomic labels do not supply independently observed behavior, caller intent or receiver-response labels. The paper's BirdCLEF+ competition dataset is a separate sound-identification task, so its classification score is not a result on this El Silencio subset and does not measure communication. This can help test a multi-species detection/data workflow, but it is not a communication or meaning benchmark.
Full overview paper, pinned annotation protocol and dataset/benchmark-fit review
Birds2026
CEB v1.1 is a large library of bird recordings with human-marked sound events: 20,826 audio files and 92,857 annotations across 256 eBird species codes and 21 vocalization types. It can help test whether software finds and sorts labeled calls in recordings. It contains no independent observations of a listener's response or ground truth about what a bird meant, so it cannot test translation. The field soundscape recordings are more relevant to passive-monitoring workflows than the separate focal Xeno-Canto recordings.
Official Zenodo v1.1 record, version notes, schema and API file inventory reviewed; annotation rows, media and model remain uninspected
Birds2025
BirdSet is a large benchmark for recognizing bird species and related acoustic labels in recordings and soundscapes. The paper reports more than 6,800 recording hours, nearly 10,000 training classes and eight strongly labeled evaluation datasets totaling more than 400 hours. The live card lists a 528,434-recording XCL snapshot at 484 GB and recommends a regional HSN configuration with 5,460 training recordings, 10,296 test labels and 21 classes at 5.92 GB. These are upstream-reported benchmark figures, not a Talk2Nature reproduction. The task is acoustic classification; it does not test whether a receiver understood a call or what it meant.
Targeted primary paper, live dataset-card and maintainer license-thread review; no data acquisition or reproduction
Birds2026
WABAD is a promising large-scale bird-acoustic reference. The publisher reports 5,047 minutes of audio with 91,931 species-level annotated vocalizations from 1,192 species, 72 recording sites, 29 locations and 13 biomes. The latest Zenodo record lists 21.3 GB. This scale could help researchers test whether a sound model detects and identifies bird vocalizations across sites. The reviewed task description concerns vocalization occurrence and species labels; it does not test a receiver's response or what a call means.
Primary publisher abstract and current Zenodo v4 metadata review; no archive acquisition or reproduction
Bottlenose dolphins2026
OpenWhistle is a timely single-species research candidate: its authors report about 114 hours of 96 kHz recordings collected over five years from five known dolphins at Dolphin Reef in Eilat, plus 8,354 expert-annotated whistles in 10 categories. The authors describe the site as a semi-natural marine area open to the Gulf of Aqaba and say human–dolphin interactions are dolphin-initiated and voluntary. The official Hugging Face cards list 31,780 unlabeled audio segments and a separate labeled set split into 5,848 training, 1,253 validation and 1,253 test examples. Published tasks detect whistle presence or classify whistle type. These are upstream descriptions, not a Talk2Nature reproduction.
Targeted paper, pinned dataset cards, official dataset license, and separate code/model rights review; metadata only
Parrots2025
This study combines social-network observations with monk-parakeet vocalizations. It reports networks for 337 birds, relatedness for 100, and 5,599 vocalizations from 229 individuals over two years. Call-diversity associations varied by year; close affiliates had more dissimilar calls. These are observational associations, not evidence that a specific call carries a known message.
Targeted article, deposit-metadata and pinned-repository review; metadata only
Birds, mammals, insects, reptiles and amphibians2024
iNatSounds is a substantial broad animal-audio resource: the paper reports about 230,000 recordings (around 1,200 hours) contributed by more than 27,000 recordists, covering over 5,500 species. The 2024 README reports 5,569 species across birds, insects, amphibians, mammals and reptiles. Its task is file-level species classification; every recording has a single species label. This makes it useful for studying acoustic recognition and pretraining, not event-level meaning, animal behavior or whether a receiver understood a call.
Targeted paper, pinned dataset README/terms, media-licensing guidance and implementation/model-release review; metadata only
Zebra finches2026
V33DA is a caller-attribution task: given an already-detected zebra-finch vocalization and the birds visible at that moment, identify the bird that produced it. The pinned card reports 33,625 verified events from 10 physically distinct birds across three experiments, with synchronized audio, video and pose; caller labels derive from on-body accelerometer vibration. In the experiment-held-out split, all 5,829 BP05 events from three callers (brown, purple, yellow) form the test set. That is a genuine unseen-bird group according to the card, but not a clean explanation of generalization because several cohort conditions shift together. This is an engineering benchmark candidate, not a communication-meaning result.
Pinned dataset-card, code and sample metadata reviewed; media uninspected
Birds of northeastern North America2026
The latest Zenodo v2 record describes 1,302 ten-minute forest soundscapes—about 217 hours—from 104 sites in four northeastern U.S. parks, with approximately 183,000 vocalization annotations. The version-pinned site_metadata.csv has 104 unique sites; recording_metadata.csv independently has the same 104 site IDs. Its site counts are ACAD 31, HBEF 24, KAWW 24 and MABI 25. The landing-page opening summary says KAWW has 20 sites, so the headline total is supported while that park breakdown remains inconsistent. The v2 README says 88 bird species, matching the publisher abstract and earlier v1; however, species_metadata.csv has 96 rows marked IsBird=Yes, all with nonzero annotation counts totaling 182,633. The species-count discrepancy therefore remains unresolved. This is a promising field-acoustics resource for event detection, localization and species classification, not a communication dataset.
Publisher abstract and rights notice, complete Zenodo v1/v2 metadata, and checksum-verified aggregate inspection of the four-file v2 metadata sample; no audio or annotation rows reviewed
Great tit (Parus major)2024
The Wytham Great Tit Song Dataset is an unusually dense wild-population resource: its authors report 109,963 annotated songs and 1,161,033 acoustic notes from more than 400 individual birds, collected in Wytham Woods, Oxfordshire, from 2020 to 2023. The complete current OSF API root lists 23 files totaling 12,184,678,510 bytes: 14 data files (12,184,652,639 bytes) and nine documentation files (25,871 bytes). The four song-archive parts total 11,380,411,540 bytes and ten metadata/map files total 804,241,099 bytes. The OSF node declares CC BY 4.0 and reports last modification on 2023-07-10; no explicit dataset version label was present in the inspected node response. The 2025 paper already trained a ViT metric-learning model on this corpus, generated 384-dimensional embeddings for all 109,963 songs, and used them to analyze repertoire similarity and demographic drivers of song culture. It then used automated clustering plus manual song categorization; the reported 0.87 validation accuracy belongs to a separate ResNet50 validation procedure. The dataset supports valuable song-structure and population-methods work, but it has no independently observed interaction context, receiver response or label for what a song means.
Primary 2024 dataset paper and 2025 full article/methods reviewed; author documentation, OSF API manifest and documented coordinate fields reviewed. On 2026-10-25, one 6,426-byte documentation file was fetched only to verify its current OSF SHA-256 and MD5; its text was not reviewed or redistributed. No audio archive acquisition or model reproduction.
Seba’s short-tailed bat2026
In ten awake Seba’s short-tailed bats, the authors report auditory-cortex firing patterns associated with whether the animal produced an echolocation pulse or a social communication call, and whether some calls occurred singly or in short sequences. Neural activity tracked the number of social-call syllables before vocal onset; the corresponding relationship for echolocation-pulse count appeared after onset. This is evidence about neural activity around the animal’s own vocal production, not a decoded message or receiver response.
Primary article full text reviewed; data/code repository metadata only
Budgerigars2026
The publisher abstract describes ten budgerigars monitored under stable housing from June 2025 through May 2026. A Raspberry Pi system processed audio locally using a pretrained YAMNet feature extractor and a task-specific binary classifier. The study defined a vocal-activity second as a one-second interval containing at least one detected vocalization and reports daily and seasonal patterns. This is prior art for long-duration, edge-based vocal-activity monitoring.
Primary publisher abstract reviewed
Zebra finches2026
V33DA++ is a companion to the physically verified zebra-finch caller-attribution dataset. Its card describes 3,784 reviewer-verified overlapping-call rows, 33,625 single-vocalizer events with plus/minus two seconds of context, and 3,784 overlapping events with the same context. The overlap_padded bucket repeats the overlap events with padding, so the provider's 41,193-row total is not 41,193 unique events. The recordings and all ten birds are the same as V33DA. The card suggests source separation, vocal-activity detection, audio-visual synchronization and active-speaker detection; caller evidence is anchored by on-body accelerometers, which are an oracle label source rather than an admissible predictor.
Pinned dataset-card and public code/split metadata reviewed; media uninspected
Zebra finch (Taeniopygia guttata)2026
Female zebra-finches changed the timing and acoustic structure of their calls during exchanges. The paper reports more than 1,000 hours of recordings and analyzes 834,309 calls from 21 birds across 40 pair interactions in Methods; one Results passage says 20 birds, so the count is inconsistent. The authors define a response as a first call within one second after a partner's call ends. They report responses after about 44% of partner calls, compared with an estimated 10% chance level, and describe birds responding to live sounds from an interactive audio model in a controlled lab study. The model predicts patterns such as which bird may call next and when; this is evidence about acoustic timing and contingency, not a translation of calls into human language.
Primary preprint and two separate Zenodo release records reviewed; no archive acquisition or reproduction
Zebra finch (Taeniopygia guttata)2025
The paper reports that 12 adult zebra finches learned to discriminate all 11 researcher-defined call types in an operant Go/No-Go task. The stimuli came from a library described as more than 8,000 call renditions by 45 vocalizers, and birds generalized to unfamiliar renditions. The paper already compares acoustic-classifier confusion with behavioral confusion and reports more errors among call types used in similar behavioral contexts than its acoustic-similarity comparison predicted. The authors' README describes the spectrogram classifier matrices as leave-one-vocalizer-out LDA results. Thus basic model-versus-bird confusion alignment is published prior art. A separate 2026 article applies BirdAVES to supplied call labels, but the reviewed article does not directly compare those modern-model errors with the measured behavioral confusion matrix.
Peer-reviewed article text and public release metadata reviewed; archive contents and analyses not reproduced
Zebra finch (Taeniopygia guttata)2026
The 2026 article introduces multi-task classification using frozen BirdAVES embeddings: it predicts 11 supplied call-type labels and six behavior-derived functional/semantic categories as an auxiliary target. The authors report a best call-type result of 87.11% accuracy and 82.76% macro-F1, versus 84.86% and 78.89% for their single-task call-type baseline. They report one predefined 70:15:15 split by bird. These are author-reported results, not reproduced by Talk2Nature. The semantic categories are supervised labels from an existing taxonomy; they are not discovered meanings or an open-ended translation.
Publisher article review plus current indexed claim check; metadata refreshed; no reproduction
Passerine birds (songbirds)2026
This is a strong example of large citizen-science sound collections enabling a useful cross-species analysis. The paper reports eight recurring acoustic motifs in songs from thousands of passerine species and associations between motif distributions, species traits and sound-propagation conditions. The author repository describes a 37-dimensional acoustic feature space built from modulation power spectra and unsupervised clustering. This is evidence about acoustic form and broad ecological patterns; the reviewed sources do not establish call meanings, receiver comprehension or a general bird-to-human language model.
PubMed abstract, authors’ public repository README and pinned tree metadata, GBIF metadata, five public Xeno-canto record-page license examples, and official API documentation reviewed; full article methods, supplement and full recording-level study inventory not reviewed; no file payloads or code acquired or run
Zebra finch (Taeniopygia guttata)2025
BirdPark is a specialized synchronized recording system combining video, stationary microphones and animal-borne accelerometer transmitters. The paper uses redundant channels to help find vocal events and separate likely caller sources in freely behaving zebra-finch groups. The public records describe several related but distinct releases: a Zenodo v2 record claiming 112 minutes across four experiments, a later annotator-training pack with a 4.5 GB pair-example archive, and an ETH record describing ten mixed-sex pairs (20 birds) with annotated clean and overlapping vocalizations. These resources could inform bounded vocal-event, overlap and caller-attribution methods. Their labels do not establish what a call means or whether a bird understood a message.
Paper abstract and selected searchable methods passages plus exact dataset-record metadata reviewed; no dataset, code or model acquired or run
Zebra finch (Taeniopygia guttata)2026
The article reports 1,454 randomly selected 44.1 kHz vocalizations from three isolated zebra finches recorded between 55 and 136 days post-hatch, with independent manual clusters from three experts for each bird. Mean ARI/NMI were 0.78/0.83 for expert-to-expert manual agreement, 0.73/0.81 for human-refined clusters and 0.62/0.71 for fully automated Whisper-based clustering. The authors also report about 90 minutes per bird for manual clustering versus 45 minutes for semi-automated proofreading. These are published results, not Talk2Nature reproductions. The method suggests that model proposals plus human correction may scale annotation while preserving expert review; its categories are acoustic clusters, not validated call meanings.
Full article and source metadata reviewed; no reproduction or third-party payload acquisition
Multiple mammals and birds; monk parakeet subset2025
The article reports 250,670 calls from 14 taxa with annotated fundamental-frequency (F0) contours. The Dryad release lists waveforms and annotations for 13 taxa; it also includes F0 predictions for bottlenose dolphins whose waveforms could not be published. The listed release totals 27.86 GB. Rodents contribute 224,705/250,670 calls (about 89.6%), while monk parakeets contribute 233. In the paper’s target-trained CREPE comparison, monk-parakeet pitch accuracy is 0.30 under that study’s half-semitone metric. This is an acoustic-feature estimation score, not a communication or meaning score. The 2025 paper cites Smith-Vidaurre et al. (2021, 2023) papers and deposits as sources for the 233 semi-automatically annotated monk-parakeet calls. The 2021 Figshare record lists 12 CSV/KML/README files and no waveform. The 2023 Figshare v2 inventory lists a 440,873-byte selection table and a 31,079,669-byte WAV archive; its README says their sound.file names correspond across those files and reports 1,596 calls in the table, 1,582 used in the paper. This offers a possible metadata-only source-file comparison, but the exact mapping of the F0 paper's 233 calls to the 2023 files remains unverified. The Dryad API v9 manifest lists 15 files totaling 27,863,357,661 bytes and identifies monk_parakeets.zip as 5,076,570 bytes with a provider-reported SHA-256. This is inventory metadata only: the archive and its file_list.txt have not been downloaded or inspected, and the checksum has not been recomputed locally.
Full-text article and source metadata reviewed; metadata-only, with no reproduction or specialist review
Insects (Orthoptera and Cicadidae)2026
InsectSet459 is a substantial, multi-species resource for insect sound identification and testing audio pipelines across varied sample rates. Its species labels do not identify behavior, caller intent, receiver response or message meaning. The recording diversity also means a phone microphone should not be assumed to capture every insect signal; high-frequency work may need a separately validated sensor.
PubMed primary abstract, official Zenodo v1.1 description and file list, and author repository root/README reviewed; full article methods and annotation CSV contents not reviewed; no media, code or model acquired or run
Companion animals2026
Among the reviewed resources, no public companion-animal corpus combines naturally occurring household audio, independently time-aligned behavior or receiver-response labels, verified household/individual/session grouping, and fully reviewed reuse terms. CatMeows describes 440 sounds from 21 cats with context and owner/session identifiers, but its CC BY-NC 4.0 terms do not provide general commercial permission and the contexts are elicited. A dog-bark paper analyzes 16,789 segments from 74 dogs during designed home exposures; its reported classifier predicts associations with those situations, not what dogs mean. Parrot surveys describe caregiver reports about human-language mimicry and names, not an acoustic training corpus. A budgerigar research deposit contains derived call features and separate social observations without source call WAVs or event-aligned behavior; a separate 2026 abstract reports local vocal-activity monitoring, not a conversation model.
Synthesis of existing source-linked records; no new data, code or model inspected
Insects2023
Dormagen et al. combined waggle-phase detection, individual tracking, follower observations and feeder records. The paper reports more than 100,000 detected waggle phases. A manually annotated subset covers 176 videos, 3,278,121 detections from 28,873 tracks and 4,389 bee IDs, including 1,107 waggle phases and 279 follower events. The Zenodo v1.0 API lists 11 deposited files totaling 4,870,358,178 bytes (4.87 decimal GB), alongside more than 46 million 2019 track detections. This is a rare data path connecting a well-defined sender signal to observed receiver behavior and known food-site outcomes.
Targeted primary-paper and dataset-metadata review; no payload acquisition or reproduction
Cattle2026
Kate and Neethirajan describe MooLogue as a Flutter mobile application for learning about cattle vocalizations. The paper reports a curated library built from 569 clips, 48 fine-grained behavioral classes and three commercial dairy farms, grouped into nine high-level learning categories. Its published v1.0 is described as an offline-first educational app with curated examples and metadata bundled for normal use. The pinned public repository README instead describes a cloud workflow that uploads recordings to AWS, runs PyTorch inference and returns a label, confidence and recommended action. The repository snapshot is dated 2025-11-25, and the paper does not establish that this exact snapshot is the implementation of the later v1.0. Treat the discrepancy as unresolved. This is a useful mobile-learning and version-control case study, not evidence that an app can translate cattle calls or provide validated welfare advice.
Targeted article and pinned-repository review; metadata-only, no reproduction
Birds2026
The NBM resource is a useful example of volunteer-supported passive-acoustic annotation for a narrow bird-monitoring task. Its training collection is reported as 2,077 audio files, 13,359 time-and-frequency vocalization annotations, about 38 hours and 117 species; 883 training files came from Xeno-Canto. The paper reports a two-stage object detector and mAP@IoU0.5 of 0.67 on 45 species. Those outputs localize and classify calls on spectrograms. They do not label observed behavior, receiver response, intention or meaning. The paper's test-set description says six XC files per species, 270 total, selected from recordings uploaded after the training collection; its detection-evaluation section instead says 271 recordings. Resolve that one-file discrepancy before attempting reproduction.
Publisher methods/evaluation and official record metadata reviewed; data, code and weights remain unacquired
Birds2026
BirdNET Live is an open-source Flutter app for Android and iOS, with Windows described as experimental. The v1.2.1 release identifies bird species from sound using on-device BirdNET+ inference; the README describes 9,789 species, Live/Point Count/Survey/File Analysis/ARU modes, reviewable sessions and local exports. The official privacy policy says the app has no backend and does not send audio to external servers. The Getting Started guide says microphone permission is required across modes and completed sessions are saved automatically for review. Three optional network features—map tiles, reverse geocoding and weather—are separately gated and off by default; the policy says requests can reveal tile coordinates or session location/time and IP address to the respective provider. Source code is MIT and the exact bundled trained weights are declared Apache-2.0 in separate files at v1.2.1. This is a strong local-first field-recording and species-identification precedent, not evidence that an animal call has been translated or that the app understands a bird’s intent.
Official app documentation/privacy policy and pinned GitHub release reviewed; no app install, model execution or independent validation
Brown-throated parakeet (Eupsittula pertinax)2021
The study analyzes one call type, loud contact calls, rather than the species’ full vocal repertoire. It reports population-level differences in call duration, frequency measures and spectrogram structure/variability across five island populations and mainland Venezuela. The authors explicitly say receiver experiments are needed to test whether these acoustic patterns correspond to receiver responses. This is a useful parrot-specific acoustic-variation study, not a call-meaning or translation result.
Targeted publisher-indexed primary-article passages and the official Dryad v4 record/file manifest reviewed; no article or dataset file acquired.
No notes match this combination. Clear the filters or try another phrase.
Talk2Nature