FIG 9.0 — Reference · FAQ

Everything we’ve been asked. Citation-grounded.

Hundreds of questions about American crow vocalizations, the AI audio foundation models that map them (BirdNET, Perch 2.0, NatureLM-audio), the methods underneath, what we can and can’t responsibly claim, and the broader field of computational bioacoustics. Each answer is anchored in primary research and links back to the source page where it was first written.

292 answers · 14 sections

Section

The Crow

13 Q

Foundational questions from the The Crow pillar pages.

Why focus on American crows specifically?

American crows (Corvus brachyrhynchos) have one of the most-studied corvid behavioral literatures, a large public CC-licensed audio corpus, and a vocal repertoire dense enough to map meaningfully. The Marzluff Seattle research program supplies the cognitive context; Mates et al. supply the acoustic baseline.

What makes the American crow worth studying as a communicative animal?

Crows recognize and remember individual human faces for years, pass face-recognition information socially across family groups, use tools, live in cooperative family groups with helpers, and demonstrate causal reasoning. The substrate for cultural communication exists; the vocal channel is one part of it.

How many types of vocalizations do American crows make?

American crows produce nine acoustically-distinct call types when grouped by self-supervised audio embedding: territorial caws, mobbing alarm, assembly calls, rattle, juvenile begging, companion calls, quiet grunts, loud grunts, and an exceptional category for atypical vocalizations. Older field literature collapsed these into 4 to 6 named types; the AI-discovered geometry is more granular.

Read more inTypes of crow vocalizations

What is a territorial caw and what does it communicate?

Territorial caws are long-duration vocalizations (typically 200-500 ms) emitted from a stable perch, often by paired adults facing outward across the territory boundary. The acoustic signature carries caller identity (recoverable via spectral profiling per Mates et al. 2014), caller sex (statistical distinguishability via fundamental frequency and call duration), and approximate intent (territorial defense versus general advertisement). American crow territorial caws differ from carrion crow and common raven equivalents in subtle spectral details that the bioacoustic literature has documented; the call type is the most studied of the species's vocal repertoire and forms the largest behavioral-context cluster in the CrowLingo atlas.

Read more inTypes of crow vocalizations

What is a mobbing call and when do crows produce it?

Mobbing calls are compressed, urgent, harmonically-rough caws delivered in rapid sequences — typically 3-8 calls per second at sustained amplitude — by individual crows or recruited groups targeting an aerial predator (commonly Cooper's hawk, red-tailed hawk, great horned owl). The vocalization is tightly packed in time, spectrally distinct from territorial calls, and serves the dual function of harassing the predator and recruiting additional conspecifics to the mobbing aggregation. The Marzluff lab's face-recognition work shows that crows also produce mobbing-like vocalizations toward learned-threat humans, with the calls persisting for years and transmitting socially to crows not present at the original encounter. The mobbing-call cluster is the second-largest behavioral context in the atlas after territorial caws.

Read more inTypes of crow vocalizations

Do American crows have dialects across different regions and family groups?

Yes, in two measurable senses. First, individual caller identity is recoverable from a single caw via spectral features (Mates et al. 2014), meaning each crow carries an acoustic signature unique enough for AI embeddings to distinguish. Second, family-group acoustic centroids — the statistical center of vocal patterns within a corvid family — differ measurably between geographically-separated populations, satisfying the technical definition of dialect. The dialect emerges from vocal learning during juvenile development; American crow, like all songbirds, has the neural machinery (HVC, RA) for learning vocalizations from conspecifics. Pacific Northwest crow populations show subtly different vocal patterns than Atlantic-coast or Midwestern populations; whether the differences rise to the level of mutual unintelligibility (the human-language threshold for dialect) is an open question the field hasn't fully settled.

Read more inTypes of crow vocalizations

What is a syrinx and how does it differ from a human larynx?

The syrinx is the bird vocal organ, located at the bottom of the trachea where the airway splits into the two bronchi — anatomically distinct from the mammalian larynx (which sits at the top of the trachea and has vocal cords). The crow syrinx has two semi-independent membrane sources, one on each bronchus, which is why some bird species can produce two notes simultaneously. The two-source architecture, controlled by tracheobronchial muscles and modulated by air pressure from the air-sac system, shapes the spectral richness of every vocalization a crow produces. Coen Elemans at the University of Southern Denmark has produced much of the modern syringeal biomechanics literature; the field's understanding of how avian sound production actually works has advanced substantially over the past decade.

What frequency range do American crow vocalizations cover?

American crow vocalizations span roughly 200 Hz to 8 kHz, with most adult-caw fundamental frequencies falling between 500-1500 Hz and harmonic energy extending well above. The species's syringeal anatomy biases the repertoire toward impulsive, harmonically-rich calls (the recognizable caw family) rather than the sustained tonal singing of warblers or thrushes. Juvenile begging vocalizations skew higher (often 1.5-3 kHz fundamental); rattle calls and contact calls cover the full range with different spectral profiles. The frequency window is well-characterized in the bioacoustic literature, with Mates et al. 2014 providing detailed acoustic profiling of identity and sex cues across the spectrum.

What is the vocal map?

The vocal map is a two-dimensional projection of a 1,024-dim NatureLM-audio embedding space. Each bright dot is a real CC-licensed crow recording; soft background points illustrate cluster geometry. Click any bright dot to play the audio, see its real spectrogram, and read the AI cluster narrative.

What is UMAP?

UMAP is uniform manifold approximation and projection — a non-linear dimensionality reducer that flattens high-dimensional embeddings to two dimensions while preserving local neighborhood structure. Used here to make a 1,024-dim audio embedding inspectable in a scatter plot.

Are the cluster boundaries discovered or assigned?

Both. HDBSCAN discovers dense regions in the full embedding; the names attached to those regions come from human biologists matching exemplars against the prior descriptive vocabulary in Marzluff & Angell, Mates et al., and Verbeek et al.

Can crows recognize human faces?

Yes. The Marzluff mask experiments at the University of Washington (2006–2014) showed American crows can recognize and remember the faces of individual humans for years, and pass the information socially across family groups — including to crows not present at the original event.

Are crows social learners?

Yes. The mask-experiment social-learning result implies information moves between crows; whatever the carrier (vocalizations, behavioral cues, observation of others' responses), there was transmission. The substrate for cultural communication exists.

Methods

11 Q

Foundational questions from the Methods pillar pages.

What's the difference between traditional bioacoustics and ALP?

Traditional bioacoustics hand-crafts features chosen by humans (mean frequency, harmonic-to-noise ratio, etc.) and trains classifiers on those features. Animal Language Processing (ALP) trains audio foundation models on millions of unlabeled recordings via self-supervised objectives, then uses the model's learned embeddings directly. The trade-off is interpretability for raw signal.

What is an audio foundation model?

An audio foundation model is a transformer trained on large unlabeled audio corpora via self-supervised objectives like masked prediction. It produces a learned vector representation of any input clip that captures information richer than hand-engineered features. Examples: BirdNET, Perch 2.0, NatureLM-audio.

What is self-supervised learning in audio?

Self-supervised learning trains a model on unlabeled audio using a pretext task — predict the masked portion of a spectrogram, or recognize that two clips come from the same source — so the model learns rich representations without anyone telling it what each clip contains. The downstream effect: a 1,024-or-1,536-dim vector that captures more than any hand-engineered feature.

Why is self-supervised audio useful for bioacoustics?

Labeled wildlife audio is scarce and expensive. Self-supervised models pretrain on the vastly larger pool of unlabeled wildlife recordings and only need labels for the final downstream task. That's how BirdNET, Perch, and NatureLM-audio all reached SOTA.

What is a latent space?

A latent space is the high-dimensional vector space a model lives in after it embeds an input. Every audio clip becomes a point; distances between points correspond to acoustic similarity; directions in the space correspond to dimensions of variation the model found important. UMAP projects that geometry down to 2D for inspection.

Why are nearest-neighbor queries done in high-dim, not 2D?

UMAP preserves local neighborhood structure for visualization but loses information in projection. Similarity search and nearest-neighbor queries always run on the full 1,024-or-1,536-dim embedding to keep the geometry honest.

What is NatureLM-audio?

NatureLM-audio is Earth Species Project's audio-language foundation model for bioacoustics, presented at ICLR 2025. It combines a BEATs audio encoder with a Llama-3.1-8B-Instruct language backbone and answers natural-language questions about audio clips zero-shot. SOTA on the BEANS-Zero benchmark.

What can NatureLM-audio actually do?

Zero-shot species classification, focal-bird identification, count estimation, life-stage classification, and natural-language captioning of arbitrary bioacoustic clips — without requiring task-specific fine-tuning. It's the first model where you can ask 'how many crows are calling' and get a useful answer.

What is the difference between BirdNET and Perch 2.0?

BirdNET is a Cornell-built EfficientNet-B0 species classifier optimized for real-time detection; it emits a 1,024-dim embedding biased toward species-discriminative features. Perch 2.0 is a 2025 Google Research model with an EfficientNet-B3 backbone, ~12M parameters, and a (5, 3, 1,536) embedding that captures finer within-species variation.

Which bioacoustic model is best for crows?

Perch 2.0 is the strongest published model for crow work in 2026: its embedding geometry captures within-species detail better than BirdNET, and unlike NatureLM-audio it returns a fixed-size vector suitable for similarity search and UMAP projection.

Is BirdNET still relevant in 2026?

Yes. BirdNET remains the dominant deployment model for long-term acoustic monitoring, phone-based citizen science (Merlin), and detection-first pipelines where Perch's compute overhead isn't justified.

Section

Decoding

4 Q

Foundational questions from the Decoding pillar pages.

What can AI actually decode about crow vocalizations?

Caller sex, individual identity, behavioral context (territorial, mobbing, recruitment, affiliative), and approximate intent — all from a single half-second of crow voice. What it cannot decode: lexical meaning, compositional syntax, or anything that would deserve the word translation.

Do crows have grammar?

Unknown. Statistical models hint at structured composition — caw-rattle sequences are non-random — but the behavioral evidence that crows treat sequence order as meaningful is thin. This is where the next five years of corvid bioacoustic research lives.

Do American crows really have regional dialects?

Yes, in two empirically-documented senses. Family-group acoustic centroids — the statistical center of vocal patterns within a corvid family — differ measurably between geographically-separated populations, with between-population differences that exceed within-population variation. Mates et al. (2014) demonstrated this with hand-crafted acoustic features; modern embedding-based pipelines using Perch 2.0 and NatureLM-audio reproduce the result at finer resolution. Dialect emerges from vocal learning during juvenile development — American crow, like all songbirds, has the HVC and RA brain regions that support learning vocal patterns from conspecifics. Whether dialect differences rise to the level of mutual unintelligibility (the human-language threshold) is unsettled; the field treats American crow dialects as real and biologically meaningful while remaining cautious about over-claiming the linguistic analogy.

Can AI identify individual crows from their vocalizations alone?

Yes, with classification accuracy approaching the noise-limit of the recording itself. Harmonic emphasis — the relative loudness of second and third harmonics versus the fundamental frequency — produces an acoustic fingerprint specific enough to distinguish individual American crows reliably. Audio embedding models (BirdNET 1,024-dim, Perch 2.0 1,536-dim) capture this individual signature implicitly in their learned vector representations; nearest-neighbor queries on a single caw can identify the specific bird among a known reference population. The capability is documented in Mates et al. 2014 and confirmed in newer embedding-based work. Individual identification from voice supports research on family-group structure, dispersal patterns, and longitudinal behavioral tracking without requiring physical banding.

Frontier

4 Q

Foundational questions from the Frontier pillar pages.

What is on the CrowLingo v1-v5 roadmap?

v0 (live): interactive atlas with 16 real recordings, AI narration, cluster pages. v1: Perch 2.0 embeddings + similarity search. v2: upload-your-own-crow with cluster placement. v3: extend to other corvid species. v4: Hugging Face dataset release. v5: real-time bidirectional synthesis — speculative end-state, no promise.

Can I contribute crow recordings?

Yes. Field recordings under CC-BY-SA license, with city-coarsened location metadata and behavioral notes when available, can be submitted via the contribute page. We don't accept recordings under restrictive Macaulay or commercial licenses, or recordings with identifiable human voices in background.

Is it ethical to do audio playback experiments with wild crows or near active nests?

Playback near active nests is effectively prohibited — reproductive stress is the highest-stakes welfare cost for wild birds, and the default position across the corvid research community is no playback during the breeding season near nest sites, with vanishingly rare exceptions requiring formal ethics-board justification and documented research-question necessity. Outside nesting contexts, playback can be ethically defensible if conducted as a pre-registered experimental protocol with observer, time-bounded sessions, immediate halt on distress signals, and disclosed reporting. CrowLingo's editorial floor declines all wild-crow playback in product features; the atlas is record-and-observe, not call-and-respond. The Demartsev et al. carrion-crow wearable-logger methodology represents the modern non-playback approach to acoustic-context research.

Why should AI bioacoustic systems never simulate or reproduce crow alarm calls?

Alarm calls function as recruitment signals — a vocalization recruiting conspecifics to mob a perceived threat. A simulated alarm triggers real recruitment from crows who cannot distinguish a synthesized stimulus from an actual predator event, producing wasted vigilance behavior, disrupted foraging, cumulative habituation that erodes the call's biological function, and stress on the entire local population. The risk compounds across deployments: an AI app that plays alarm calls 'just to listen to crows' in a residential neighborhood produces measurable population-level harm. This is the most-cited ethical floor across modern corvid research and the field's most-consistent position; CrowLingo's atlas reproduces the principle in product design (no playback features anywhere on the site).

The Crow · Vocal clusters

9 Q

Per-cluster questions covering each of the nine acoustically-distinct American crow call types in the vocal atlas: territorial caws, mobbing alarm, assembly calls, rattle, juvenile begging, companion contact, quiet grunts, loud grunts, and atypical vocalizations.

What is the territorial caw cluster in the CrowLingo vocal atlas?

Long-duration caws emitted from a perch, often paired and faced outward across a territory boundary. Carries caller identity, sex, and approximate intent.

What is the mobbing alarm cluster in the CrowLingo vocal atlas?

Compressed, urgent caws delivered in rapid sequences, often by recruited pairs or groups targeting an aerial predator. Spectrally rough; tightly packed in time.

What is the assembly cluster in the CrowLingo vocal atlas?

Loud, far-carrying calls that summon group members to a roost or food source. Acoustically distinct from territorial in rate and inter-call interval.

What is the rattle complex cluster in the CrowLingo vocal atlas?

Mechanical rattling vocalizations with weak harmonic structure. Frequent in affiliative and recruitment contexts. Highly individual.

What is the quiet grunts cluster in the CrowLingo vocal atlas?

Low-amplitude, close-range grunts used in affiliative and parent-offspring contexts. Subtle; only the new wearable-logger studies recover them at scale.

What is the loud grunts cluster in the CrowLingo vocal atlas?

Higher-amplitude grunts during foraging and recruitment. Often serial and quickly answered by group members.

What is the begging cluster in the CrowLingo vocal atlas?

Higher-frequency, narrower-band calls from juveniles soliciting feeding. Diagnostic spectral signature; tightly clustered.

What is the exceptional cluster in the CrowLingo vocal atlas?

Rare, atypical, or unusual vocalizations that don't fit the named categories. Where future repertoire expansion lives.

What is the companion cluster in the CrowLingo vocal atlas?

Soft contact calls between paired adults. Heavily individual and pair-specific; acoustic signatures distinguish couples.

Journal · The Crow

102 Q

Editorial questions about American crow species biology, cognition, social structure, and behavior — from the long-form journal.

Can crows actually recognize individual human faces?

Yes. The Marzluff mask experiments at the University of Washington from 2006 to 2014 showed American crows can recognize and remember the faces of individual humans for years — and pass that information socially across family groups, including to crows not present at the original event.

Read more inWhy crows recognize human faces — the Marzluff experimentscrow face recognition · Marzluff mask experiments · crow memory · corvid cognition

How long do crows remember a face?

At least several years, based on the published follow-ups of the Marzluff mask experiments. Crows continued to scold the threatening mask years after the original trapping event.

Read more inWhy crows recognize human faces — the Marzluff experimentscrow face recognition · Marzluff mask experiments · crow memory · corvid cognition

Are the Marzluff mask experiments peer reviewed?

Yes. The findings have been published in multiple peer-reviewed papers from Marzluff's lab, and replicated in spirit in adjacent corvid species. The 2005 trade book In the Company of Crows and Ravens, with Tony Angell, summarizes the broader research program.

Read more inWhy crows recognize human faces — the Marzluff experimentscrow face recognition · Marzluff mask experiments · crow memory · corvid cognition

How many call types do American crows have?

Nine, when grouped by self-supervised audio embedding: territorial caws, mobbing alarm, assembly calls, rattle, juvenile begging, companion calls, quiet grunts, loud grunts, and an exceptional category for atypical vocalizations. Older field literature collapsed these into 4-6 named types.

Read more inThe nine emergent clusters of American crow communicationcrow call types · American crow vocalizations · vocal repertoire UMAP · crow communication categories

What is the rattle call?

A mechanical, weakly harmonic vocalization unique to American crows in the corvid family. Rattles appear in affiliative, recruitment, and occasionally territorial contexts and are highly individual. Pairs and family groups use them to signal something closer to mood than position.

Read more inThe nine emergent clusters of American crow communicationcrow call types · American crow vocalizations · vocal repertoire UMAP · crow communication categories

How are crow vocalization clusters identified?

By embedding every clip with a self-supervised audio foundation model (NatureLM-audio or Perch 2.0), then running HDBSCAN — hierarchical density-based spatial clustering — on the full high-dimensional vectors. Cluster names are assigned post-hoc by listening to exemplars.

Read more inThe nine emergent clusters of American crow communicationcrow call types · American crow vocalizations · vocal repertoire UMAP · crow communication categories

Why is the American crow the right species for AI bioacoustics?

It clears a specific set of empirical bars: cognitive sophistication (individual face recognition with social transmission, demonstrated by Marzluff); cooperative breeding social structure (calls addressed to known individuals in stable networks); acoustic richness (200 Hz to 8 kHz range, 9 emergent cluster types); availability of CC-licensed audio; and a deep behavioral-observation literature anchoring cluster interpretations.

Read more inWhy the American crow is the model species for AI bioacousticsAmerican crow research · model species bioacoustics · Corvus brachyrhynchos AI · why study crows

Are crows social learners?

Yes. The Marzluff mask experiments showed face-recognition information moves between crows — including to birds not present at the original event. This establishes the substrate for cultural information transfer, which is a precondition for the more ambitious claims about meaningful crow communication.

Read more inWhy the American crow is the model species for AI bioacousticsAmerican crow research · model species bioacoustics · Corvus brachyrhynchos AI · why study crows

Could the same AI methods work on other birds?

Yes, with caveats. The embedding methods generalize. The cluster-level behavioral interpretations depend on having synchronized behavioral data, which American crows have unusually deep, and most other species don't.

Read more inWhy the American crow is the model species for AI bioacousticsAmerican crow research · model species bioacoustics · Corvus brachyrhynchos AI · why study crows

Are crows intelligent?

By every operational measure researchers have applied — individual recognition, social transmission of information, episodic memory, tool use in some species, problem-solving — yes. The 'corvid intelligence' framing is supported by fifty years of accumulating evidence from independent research programs.

Read more inCorvid cognition: fifty years of breakthroughscorvid cognition · crow intelligence research · raven intelligence · Bernd Heinrich

What is the most famous crow cognition study?

The Marzluff mask experiments (2006-2014) showing American crows recognize and remember individual human faces for years and pass that information socially across family groups. The methodological rigor — control masks, multi-year follow-up, social-learning extensions — made the findings hard to dismiss.

Read more inCorvid cognition: fifty years of breakthroughscorvid cognition · crow intelligence research · raven intelligence · Bernd Heinrich

Do crows really use tools?

New Caledonian crows use tools extensively, including shaped hooked tools and tool sequences. American crows show more modest tool use in laboratory and observational settings, but the genus-level cognitive potential is established. Tool use is one of several capacities that raised the cognitive ceiling for the family.

Read more inCorvid cognition: fifty years of breakthroughscorvid cognition · crow intelligence research · raven intelligence · Bernd Heinrich

Do crows really hold funerals for their dead?

No, not in the human sense of commemorative ritual. Crows do reliably gather around dead conspecifics and produce mobbing-style alarm calls, but the careful research (Kaeli Swift's experiments in the Marzluff lab) interprets this as a threat-response behavior — treating dead crows as a 'place is dangerous' signal — rather than as grief or mourning ritual.

Read more inCrow funerals and crow grief: separating evidence from anthropomorphismcrow funerals · do crows grieve · crow death behavior · Kaeli Swift crow research

Do crows grieve?

Unknown. The threat-response evidence is strong; the grief evidence is thin. Grief in the empirically meaningful sense would require physiological measurement of stress hormones, prolonged behavioral changes in surviving family members, or comparative work that hasn't been done at the relevant scale. The question remains open.

Read more inCrow funerals and crow grief: separating evidence from anthropomorphismcrow funerals · do crows grieve · crow death behavior · Kaeli Swift crow research

Who studied crow funeral behavior scientifically?

Kaeli Swift's PhD research at the University of Washington (in John Marzluff's lab) characterized American crow responses to dead conspecifics in controlled experiments. Subsequent work has extended the findings to related corvid species.

Read more inCrow funerals and crow grief: separating evidence from anthropomorphismcrow funerals · do crows grieve · crow death behavior · Kaeli Swift crow research

Do American crows use tools?

Occasionally and opportunistically in the wild, more reliably in controlled laboratory settings. They don't show the species-typical, multi-step tool culture that New Caledonian crows demonstrate. The cognitive capacity is probably similar; the expressed behavior differs ecologically.

Read more inCrow tools and the cognitive ceilingNew Caledonian crow tools · crow tool use · Kacelnik Oxford corvids · Gavin Hunt tool research

What corvid species is famous for tools?

The New Caledonian crow (Corvus moneduloides). Documented by Gavin Hunt, Russell Gray, Alex Kacelnik and colleagues from the 1990s onward. Shapes hooked tools from twigs and pandanus leaves, exhibits local 'tool cultures' across populations of New Caledonia.

Read more inCrow tools and the cognitive ceilingNew Caledonian crow tools · crow tool use · Kacelnik Oxford corvids · Gavin Hunt tool research

Why does tool use matter for crow vocal research?

Because tool use is one of several capacities that establish the cognitive ceiling of the family. If corvids can operate in a cognitive space that includes means-end reasoning and material culture, their vocalizations are happening in a context substantially richer than reflexive signaling — which doesn't prove vocal meaning is rich, but raises the plausibility of investigation.

Read more inCrow tools and the cognitive ceilingNew Caledonian crow tools · crow tool use · Kacelnik Oxford corvids · Gavin Hunt tool research

Who is Bernd Heinrich?

A biologist (University of Vermont, earlier UC Berkeley) who spent several decades observing wild common ravens. Author of Ravens in Winter (1989) and Mind of the Raven (1999), plus many peer-reviewed papers. The naturalist methodology — patient field observation across years — anchored much of what corvid cognitive research now takes as established baseline.

Read more inBernd Heinrich, Mind of the Raven, and the slow-naturalist traditionBernd Heinrich Mind of Raven · naturalist observation tradition · slow science animal behavior · common raven research

What did Heinrich establish about ravens?

Individual recognition across seasons, cache-and-retrieval at long delays, cooperative anti-predator behavior, play behavior with no immediate function, complex social structures. Each of these is now part of the corvid-cognition canonical literature; many were first richly characterized by Heinrich's patient field observation.

Are Heinrich's findings still relevant in the AI era?

Yes. The AI methods can characterize vocal repertoires at scale but they need contextual ground truth — what the animal was doing, who it was with — that naturalist work produces. The field needs both: AI for scale, naturalist work for meaning. Heinrich's tradition is the methodological conscience of contemporary research.

What does it mean when crows are cawing loudly?

Depends on the pattern. Paced single-bird caws from a perch are usually territorial advertisement — the bird is announcing its presence to neighbors. Rapid-fire multi-bird caws are usually mobbing — there's a predator nearby and the crows are recruiting each other to harass it. The rate difference is the cleanest distinction in the repertoire.

Read more inWhat you'll hear in your backyard: a beginner's guide to crow vocalizationsbackyard crow sounds · what do crow calls mean · crow vocalizations beginner · American crow calls explained

Why do crows make rattle sounds?

Rattle calls appear in affiliative contexts (between paired or familiar individuals), sometimes in recruitment, occasionally during territorial encounters. The acoustic structure is mechanical and weakly tonal — closer to a small wooden toy being shaken than to a typical bird call. American crows are one of the only corvids producing this exact acoustic structure.

Read more inWhat you'll hear in your backyard: a beginner's guide to crow vocalizationsbackyard crow sounds · what do crow calls mean · crow vocalizations beginner · American crow calls explained

What time of day are crows loudest?

Dawn chorus (the first hour after sunrise during breeding season) and the pre-roost gathering hour before sunset. The dawn pattern is dominated by territorial advertisement; the dusk pattern is dominated by assembly calls to roost sites.

Read more inWhat you'll hear in your backyard: a beginner's guide to crow vocalizationsbackyard crow sounds · what do crow calls mean · crow vocalizations beginner · American crow calls explained

How can I tell if I'm hearing a crow or a raven?

Listen to the fundamental frequency: crow caws are typically 500-1,500 Hz (bright caw), raven calls are typically 200-600 Hz (deeper, rougher gronk). The raven sounds croak-like; the crow sounds clear and tonal. Habitat also helps — ravens are rare in urban areas east of the Rockies, where crows dominate.

Read more inHow to tell a crow from a raven by soundcrow vs raven sound · tell a crow from a raven · American crow common raven · corvid identification by sound

Do crows and ravens make the same sounds?

No. They share the corvid acoustic toolkit broadly but differ in fundamental frequency range, tonal versus noisy texture, and specific repertoire elements. American crows have a distinctive mechanical rattle call that ravens don't produce; ravens have a richer set of quorks and knocks documented by Bernd Heinrich's field work.

Read more inHow to tell a crow from a raven by soundcrow vs raven sound · tell a crow from a raven · American crow common raven · corvid identification by sound

Where do you find ravens vs crows in North America?

Crows dominate urban and suburban habitats across most of the continent. Ravens are more common in western mountains, boreal forests, the Pacific Northwest, desert Southwest, and Northeast spruce-fir forests. Both can occur in the same area, but the urban-versus-wilderness habitat split is the most reliable broad heuristic.

Read more inHow to tell a crow from a raven by soundcrow vs raven sound · tell a crow from a raven · American crow common raven · corvid identification by sound

Why do crows gather at dusk?

Three reinforcing reasons: thermoregulation (bodies in a tree generate warmth), predator protection (many eyes detect threats, dilution effect reduces individual risk), and information sharing (birds without successful daytime foraging follow informed individuals the next morning). The combination of all three explains why communal roosting is so consistent across the species.

Read more inWhy crows gather at dusk: the roost-formation phenomenonwhy crows gather at dusk · crow roost · urban crow roost · communal roost birds

How many crows are in a typical winter roost?

Anywhere from a few hundred (suburban-park roost) to tens of thousands (the Auburn, NY roost; Minneapolis; multiple Mid-Atlantic metropolitan roosts). The largest documented urban roosts exceed twenty thousand individuals.

Read more inWhy crows gather at dusk: the roost-formation phenomenonwhy crows gather at dusk · crow roost · urban crow roost · communal roost birds

Are crow roosts noisy?

Extremely. Pre-roost staging and roost formation produce one of the largest acoustic events in urban ecology. The combined vocalizations of thousands of birds at close range can exceed eighty decibels at the base of the roost tree. The volume drops rapidly after settling but resumes at dawn departure.

Read more inWhy crows gather at dusk: the roost-formation phenomenonwhy crows gather at dusk · crow roost · urban crow roost · communal roost birds

Are crows cooperative breeders?

American crows yes. Cooperative breeding means individuals other than the breeding pair help raise the breeders' offspring. In American crows, helpers are typically previous-years' offspring of the breeding pair who stay with the natal family for two to seven years before dispersing to establish their own territories.

Read more inThe cooperative-breeding family structure of American crowscooperative breeding crows · crow family structure · American crow helpers at the nest · delayed dispersal birds

How long do young crows stay with their parents?

Range is one to seven years, with multi-year helping being the typical pattern in dense crow populations. The decision depends on territory availability, parental survival, helper-to-breeder ratios, and other factors. In sparser populations with more available territory, earlier dispersal is more common.

Read more inThe cooperative-breeding family structure of American crowscooperative breeding crows · crow family structure · American crow helpers at the nest · delayed dispersal birds

Why does cooperative breeding matter for crow communication?

Three reasons: audience is specific and persistent (calls are addressed to known individuals in stable multi-year networks); helper-breeder coordination requires call types absent in less-social species; individual signature and family-level dialect both become functionally important because the same individuals encounter each other repeatedly across years.

Read more inThe cooperative-breeding family structure of American crowscooperative breeding crows · crow family structure · American crow helpers at the nest · delayed dispersal birds

Do American crows really mate for life?

Mostly yes. Long-term field studies document pair bonds that persist year-over-year for as long as both partners survive. Documented 'divorce' (both partners alive, pair dissolves) happens but is rare. The species is best described as 'serially monogamous with strong pair persistence' — close to 'mates for life' but not literally universal.

Read more inDo crows mate for life? What the long-term pair-bond research showsdo crows mate for life · crow pair bond · American crow monogamy · lifetime pair bond birds

Why are crow pair bonds so durable?

Three reinforcing factors: the cooperative-breeding social system requires stable pair bonds (helpers depend on breeder stability), territories are long-term assets benefiting from co-defense, and acoustic individual familiarity makes established partnerships more efficient than new pairings.

Read more inDo crows mate for life? What the long-term pair-bond research showsdo crows mate for life · crow pair bond · American crow monogamy · lifetime pair bond birds

Do crow pairs ever break up?

Yes, but rarely. Documented divorce correlates with repeated breeding failure, severe territorial disruption, or helper-takeover dynamics. The default expectation in a healthy breeding pair is multi-year persistence; dissolution is the exception requiring explanation.

Read more inDo crows mate for life? What the long-term pair-bond research showsdo crows mate for life · crow pair bond · American crow monogamy · lifetime pair bond birds

What does the Wright laboratory study?

Tim Wright's lab at New Mexico State University focuses on vocal learning, individual signature, and dyadic vocal coordination in corvids and parrots. The lab uses behavioral experimental designs to test whether claims about vocal signaling hold up — which makes their work a methodological anchor for AI-derived bioacoustic findings.

Read more inWhat the Wright laboratory is up to: contemporary corvid vocal researchWright laboratory corvid · Tim Wright New Mexico · dyadic corvid vocalizations · behavioral bird research

Why do AI bioacoustic papers cite the Wright lab?

Because behavioral validation matters. AI methods can recover statistical patterns in acoustic data; whether those patterns reflect what the animals actually do requires behavioral testing. The Wright lab's work supplies that behavioral anchor, especially for individual-signature and dyadic-coordination claims.

Read more inWhat the Wright laboratory is up to: contemporary corvid vocal researchWright laboratory corvid · Tim Wright New Mexico · dyadic corvid vocalizations · behavioral bird research

What is dyadic vocal coordination?

Vocal patterns specific to pairs of birds that interact regularly — measurable acoustic centroids or coordination patterns that distinguish a given pair from other pairs in the same population. The Wright lab has documented this in multiple corvid and parrot species; it persists across seasonal changes in habitat and food availability.

Read more inWhat the Wright laboratory is up to: contemporary corvid vocal researchWright laboratory corvid · Tim Wright New Mexico · dyadic corvid vocalizations · behavioral bird research

Are crows as smart as seven-year-old children?

The comparison traces to specific water-displacement experiments showing New Caledonian crows can solve Aesop's-fable problems at roughly child-equivalent levels — but only for physical-cognition problem-solving, not language or abstract reasoning. Generalizing from 'comparable on this task' to 'as smart overall' is a popular-coverage move, not a scientific finding.

Read more inHow smart are crows? A measured assessmenthow smart are crows · crow intelligence · are crows as smart as · corvid intelligence research

What cognitive abilities do crows have?

Individual recognition of humans and conspecifics, social transmission of information, episodic-like memory, problem-solving, tool use (modest in American crows, extensive in New Caledonian crows), numerical competence in some species. The combined evidence establishes the corvid family as cognitively sophisticated by every operational measure researchers have applied.

Read more inHow smart are crows? A measured assessmenthow smart are crows · crow intelligence · are crows as smart as · corvid intelligence research

How big is a crow's brain?

About 9 grams in an American crow (~450 grams body weight). Brain-to-body weight ratio is high for birds, comparable to or exceeding many small primates. The nidopallium caudolaterale — the avian functional analogue of mammalian prefrontal cortex — is unusually developed in corvids, supporting the cognitive flexibility researchers measure.

Read more inHow smart are crows? A measured assessmenthow smart are crows · crow intelligence · are crows as smart as · corvid intelligence research

Do crows really play?

Yes, by Marc Bekoff's standard definition of animal play (voluntary, intrinsically motivated, structurally varied, observed in safe contexts). Documented examples include ravens sliding down snowbanks, crows dropping and catching sticks mid-fall, hawk-stooping flight maneuvers in non-threat contexts, and wind-riding behavior with no foraging function.

Read more inThe unexpected richness of crow play behaviorcrow play behavior · do crows play · animal play research · Bekoff play definition

Why do crows play in adulthood?

Most non-human animal play is juvenile. Corvids — crows, ravens, magpies, jays — are unusual in maintaining substantial play into adulthood. Plausible functions: motor-skill maintenance, social bonding, and cognitive flexibility. The persistence of adult play is a marker of cognitive sophistication in the family.

Read more inThe unexpected richness of crow play behaviorcrow play behavior · do crows play · animal play research · Bekoff play definition

Do crows make play vocalizations?

Possibly, but undertested. Play vocalizations are documented in some primates and carnivores as distinct call types that signal 'I'm playing' to play partners. Whether American crows have analogous vocalizations as a distinct functional category isn't yet established with rigorous evidence.

Read more inThe unexpected richness of crow play behaviorcrow play behavior · do crows play · animal play research · Bekoff play definition

Do crows really give gifts to humans who feed them?

Some do, occasionally. The behavior is documented (Gabi Mann's Seattle case popularized it; other cases have been reported in adjacent contexts). It's not universal across all crows that receive food, and the most parsimonious explanation is that crows commonly handle and abandon small objects rather than necessarily performing a communicative 'gift' behavior. Some individual cases may involve more intentional behavior; the evidence is suggestive, not conclusive.

Read more inDo crows really bring gifts? Untangling the evidencedo crows bring gifts · crow gift giving · crows leave shiny objects · feeding crows gift

How do you befriend a crow?

Feed unsalted unsweetened peanuts (in shell), suet, or hard-boiled eggs at a consistent location and time. Don't make sudden movements. Don't try to touch them. Build trust over weeks-to-months. The Marzluff face-recognition work shows crows reliably identify and remember the humans who feed them; the relationship is real even if the 'gift-giving' part isn't universal.

Read more inDo crows really bring gifts? Untangling the evidencedo crows bring gifts · crow gift giving · crows leave shiny objects · feeding crows gift

Why do crows leave shiny objects?

Most likely because crows commonly handle and cache small objects as part of normal foraging and play behavior, and may abandon or drop objects at human-feeding sites where they've spent time. Whether some individual crows leave objects more intentionally as relationship-maintenance behavior is plausible but not rigorously demonstrated.

Read more inDo crows really bring gifts? Untangling the evidencedo crows bring gifts · crow gift giving · crows leave shiny objects · feeding crows gift

How long do American crows live?

On average six to eight years in the wild for birds that survive their first year. Maximum documented lifespan in wild populations is around twenty years (Marzluff's Seattle research has recorded several individuals at that age). Captive birds in controlled environments can live longer, sometimes over thirty years.

Read more inHow long do crows live? The longevity question, answered carefullyhow long do crows live · American crow lifespan · crow longevity · wild crow age

Why is there a big gap between average and maximum crow lifespan?

First-year mortality is very high in wild crows — most don't survive twelve months. Of those that do, average mortality is moderate but lifetime is still capped by environmental factors: collisions, predation, disease, weather. The survivors who avoid these reach old age but represent a small fraction of all hatched birds.

Read more inHow long do crows live? The longevity question, answered carefullyhow long do crows live · American crow lifespan · crow longevity · wild crow age

Did West Nile affect crow populations?

Substantially. American crow populations declined by fifty percent or more in some North American regions during the early-2000s West Nile virus outbreak. Crows have lower resistance than many adjacent species. Populations have partially recovered through the 2010s and 2020s, with regional variation in pace and extent.

Read more inHow long do crows live? The longevity question, answered carefullyhow long do crows live · American crow lifespan · crow longevity · wild crow age

Are American crows and carrion crows the same species?

No, they're separate species within the genus Corvus. American crow (Corvus brachyrhynchos) is found in North America; carrion crow (Corvus corone) is found across western Europe and parts of Asia. They occupy similar ecological niches and look very similar, but the species split is established.

Read more inCarrion crow versus American crow: a comparative species notecarrion crow vs American crow · Corvus corone Corvus brachyrhynchos · European crow research · comparative crow species

Can findings from carrion crow research apply to American crows?

In broad qualitative terms, often yes. Both species share cooperative-breeding social structure, omnivorous diet, similar cognitive baselines, and similar vocal repertoire breadth. Specific quantitative findings (call frequencies, behavioral-context probabilities) require cross-species validation before being applied to American crows.

Read more inCarrion crow versus American crow: a comparative species notecarrion crow vs American crow · Corvus corone Corvus brachyrhynchos · European crow research · comparative crow species

What's the difference between a carrion crow and a hooded crow?

Hooded crow (Corvus cornix) was historically considered a subspecies of carrion crow but is now usually treated as a separate species. Hooded crows have grey body plumage with black head and wings, easily distinguishable visually. The carrion-hooded boundary in central Europe is a well-studied natural hybrid zone.

Read more inCarrion crow versus American crow: a comparative species notecarrion crow vs American crow · Corvus corone Corvus brachyrhynchos · European crow research · comparative crow species

Are crows vocal learners?

Yes. As songbirds (order Passeriformes), crows inherit the vocal-learning machinery of their lineage. They produce calls shaped by developmental auditory experience rather than purely innate templates — which is what enables individual signatures, dialect, and the cultural transmission patterns documented in American crow populations.

Read more inWhy corvids and parrots both vocal-learn: a convergent evolutionary puzzlevocal learning corvids parrots · convergent evolution vocal learning · Erich Jarvis vocal learning · bird song learning

What animals can learn vocalizations?

A small set: humans, cetaceans, bats (some species), elephants (debated), and three bird groups — songbirds, parrots, hummingbirds. Vocal learning evolved independently multiple times across these lineages, making it a convergent-evolution puzzle for evolutionary biology.

Why does vocal learning matter for animal communication research?

Because it's the cognitive substrate that supports culturally-transmitted dialect, stable individual signatures, sequence flexibility, and the kind of communicative phenomena that more-than-reflexive signaling depends on. The vocal-learning species (corvids, parrots, cetaceans) are where the contemporary AI bioacoustics research focuses.

Are urban crows smarter than rural crows?

The evidence suggests urban crow populations show more frequent tool use and behavioral innovation than rural populations. Whether this reflects genuine intelligence differences or simply that urban crows are easier to observe (so we see more of their cognitive behavior) is a methodological caveat the research field takes seriously. Current best guess: there's a real urban advantage in flexible problem-solving, but it's partly an observation artifact.

Read more inUrban versus rural crow behavior: what the acoustic record showsurban crow behavior · city crows vs country crows · urban bird adaptation · American crow urbanization

How do urban crows respond to city noise?

Several measurable adjustments. Call pitch rises to avoid low-frequency anthropogenic noise (traffic, HVAC). Call amplitude increases. Some call types shift spectral structure to be more distinguishable from urban background noise. These adjustments are documented across many urban bird species, not just crows.

Read more inUrban versus rural crow behavior: what the acoustic record showsurban crow behavior · city crows vs country crows · urban bird adaptation · American crow urbanization

Why are there so many crows in cities?

Abundant food (garbage, pet food, agricultural runoff), reduced predator pressure (no eagles or great horned owls in most urban cores), warmer microclimates (urban heat island effect), and increased nest substrate availability. Cities are ecological windfall environments for crows, and the species exploits them effectively.

Read more inUrban versus rural crow behavior: what the acoustic record showsurban crow behavior · city crows vs country crows · urban bird adaptation · American crow urbanization

Do crows hold funerals?

Crows gather and vocalize intensely around dead conspecifics in a behavioral pattern that resembles mourning from a human perspective. The systematic research by Kaeli Swift shows this is real and species-typical. Whether the behavior reflects 'mourning' in the human emotional sense, or learning-from-death (acquiring predator information) and threat-assessment behavior, remains an interpretive question the research doesn't directly resolve.

Read more inKaeli Swift and the crow funeral questionKaeli Swift crow funeral · crow funeral research · American crow death response · do crows hold funerals

Who is Kaeli Swift?

A corvid researcher whose doctoral work at the University of Washington documented the crow funeral phenomenon under controlled experimental conditions. Her work, with John Marzluff, established that the behavior is real, repeatable, and includes learned-threat associations with the humans involved. She writes the Corvid Research blog (corvidresearch.blog) since 2014.

Read more inKaeli Swift and the crow funeral questionKaeli Swift crow funeral · crow funeral research · American crow death response · do crows hold funerals

What's the functional explanation for crow funeral behavior?

Several non-mutually-exclusive interpretations: learning-from-conspecific-death (acquiring information about predator types and locations), social information transfer (alerting others to localized danger), threat-assessment behavior (treating dead-conspecific as evidence of nearby risk). The 'mourning' framing isn't directly supported by the data and is mostly human projection.

Read more inKaeli Swift and the crow funeral questionKaeli Swift crow funeral · crow funeral research · American crow death response · do crows hold funerals

How many American crows did West Nile virus kill?

Estimates suggest the species-level population declined approximately 40% across the lower 48 states at the early-2000s trough, with regional declines as steep as 50-70% in the most affected Northeast and Midwest areas. The total mortality is in the millions of individual crows. The species has partially recovered, with regional variation in recovery extent.

Read more inWest Nile virus and the American crow population collapseWest Nile virus crows · American crow population decline · WNV bird mortality · corvid disease ecology

Why are crows especially vulnerable to West Nile virus?

American crows show near-universal mortality (close to 100%) in experimental WNV infection studies, far higher than most other passerine species. The reasons combine weaker neutralizing antibody response than other birds and behavioral factors including large communal roosts that may facilitate mosquito-mediated transmission.

Read more inWest Nile virus and the American crow population collapseWest Nile virus crows · American crow population decline · WNV bird mortality · corvid disease ecology

Have American crow populations recovered from West Nile virus?

Partially and unevenly. Northeast and Midwest populations are largely rebounded to near pre-WNV baselines. West Coast populations were less affected to begin with. The recovery is driven by decreasing virus virulence, increasing population immunity, and the species's high reproductive output. Recovery is not complete in all regions.

Read more inWest Nile virus and the American crow population collapseWest Nile virus crows · American crow population decline · WNV bird mortality · corvid disease ecology

Can crows smell?

Yes, more than the older ornithology literature credited them with. The behavioral evidence shows corvids respond to olfactory cues in measurable ways, particularly for cache recovery, scavenging, and possibly conspecific recognition. The smell sense isn't as dominant as in mammals like dogs, but it's a real sensory mode that contributes to crow behavior.

Read more inCrow olfaction: the quiet sensory modecrow olfaction · do birds smell · corvid sense of smell · bird olfactory bulb

Why did people think birds couldn't smell?

Bird olfactory bulbs are relatively small compared to mammalian olfactory bulbs at equivalent body sizes. The inference that small bulbs mean weak smell turned out to be a flawed proxy — bird olfactory architectures differ from mammalian ones rather than being simply scaled-down versions. Updated behavioral evidence across the last twenty years has corrected the picture.

Read more inCrow olfaction: the quiet sensory modecrow olfaction · do birds smell · corvid sense of smell · bird olfactory bulb

Does crow communication include smell signals?

Possibly, though much less well-studied than vocal and visual communication. Most crow communication research focuses on acoustic signals because microphones are cheap, ubiquitous, and produce data that machine learning can consume. The underlying communicative situation may include multiple channels (sound + sight + possibly smell) that an acoustic-only record captures incompletely.

Read more inCrow olfaction: the quiet sensory modecrow olfaction · do birds smell · corvid sense of smell · bird olfactory bulb

Are American crow populations growing because of climate change?

The species has expanded its range northward and to higher elevations over recent decades, with climate warming being one of several enabling factors (alongside urbanization and agricultural land use). American crow is a generalist with traits favoring expansion under changing conditions, and it appears to be a 'climate winner' in the species-level terminology used for these assessments.

Read more inClimate change and corvid range shifts: what the data showsclimate change bird ranges · American crow expansion · corvid climate adaptation · bird range shift warming

What happens to the species crows now compete with or prey on?

American crow expansion creates new ecological dynamics with the species in newly-shared ranges. Crows are nest predators that can affect songbird populations. They compete with other corvids (jays, magpies) and generalist omnivores. The community-level consequences of one species's expansion can be losses elsewhere, and tracking these effects is part of climate-change ecological research.

Read more inClimate change and corvid range shifts: what the data showsclimate change bird ranges · American crow expansion · corvid climate adaptation · bird range shift warming

Are crow vocalizations changing as their range shifts?

Mostly unstudied. Plausible changes include seasonal timing shifts (earlier spring vocalization as breeding seasons advance), foraging-call shifts as diet changes, and dialect mixing as expanding populations meet previously-isolated populations. The acoustic record from citizen-science sources could support such analysis but the systematic comparative work hasn't been done yet.

Read more inClimate change and corvid range shifts: what the data showsclimate change bird ranges · American crow expansion · corvid climate adaptation · bird range shift warming

Who is John Marzluff?

Professor at the University of Washington's School of Environmental and Forest Sciences since 1997. His research program on American crow cognition has produced the foundational modern findings on corvid face recognition, social information transfer of learned threat, and neural correlates of fear response. Author of multiple academic papers and the popular-science book Gifts of the Crow (2012, with Tony Angell).

Read more inJohn Marzluff and the Seattle crow labJohn Marzluff · University of Washington crow research · American crow cognition · Seattle corvid lab

What did Marzluff's face-recognition research show?

American crows recognize specific human faces and treat the humans associated with negative experiences (e.g., capture-and-band operations) as learned threats. The behavior persists for years. The behavior transmits to crows who weren't present at the original capture, establishing both individual face recognition and social-information-transfer of threat learning. Foundational findings published starting around 2009.

Read more inJohn Marzluff and the Seattle crow labJohn Marzluff · University of Washington crow research · American crow cognition · Seattle corvid lab

How does Marzluff lab research relate to CrowLingo?

The behavioral-context categories CrowLingo uses (territorial, mobbing, assembly, alarm, contact, juvenile-begging) are grounded in the Marzluff-tradition behavioral ecology literature. The atlas's epistemic discipline — describing behavioral-context probabilities rather than claiming definitive meanings — reflects the careful framing the Marzluff lab has modeled across its work.

Read more inJohn Marzluff and the Seattle crow labJohn Marzluff · University of Washington crow research · American crow cognition · Seattle corvid lab

Who is Bernd Heinrich?

American naturalist and behavioral ecologist born in Germany in 1940. Faculty at University of Vermont until 2002. Author of foundational books on corvid behavior including Ravens in Winter (1989) and Mind of the Raven (1999). His decades of field observation of common ravens in Maine has been one of the major influences on the modern conceptual framework for thinking about corvid cognition.

Read more inBernd Heinrich and the ravens of winterBernd Heinrich ravens · Mind of the Raven · Ravens in Winter · raven cognition research

Is Heinrich's research still considered authoritative?

His natural-history work remains foundational reference material for thinking about corvid behavior in ecological context. Modern experimental cognition work (Marzluff, Clayton, von Bayern, Pepperberg) builds on conceptual ground Heinrich helped establish. The field benefits from both approaches; Heinrich's natural-history methodology catches behaviors that lab settings miss, while modern experimental work produces stronger causal inferences.

Read more inBernd Heinrich and the ravens of winterBernd Heinrich ravens · Mind of the Raven · Ravens in Winter · raven cognition research

Are ravens and crows the same thing?

Both in the genus Corvus but separate species. Common raven (Corvus corax) is the species Heinrich primarily studied. American crow (Corvus brachyrhynchos) is CrowLingo's focal species. Ravens are larger, occupy different ecological niches in many overlapping ranges, and have somewhat different social structures. Behavioral findings from one species transfer broadly to the other but specific details require species-specific verification.

Read more inBernd Heinrich and the ravens of winterBernd Heinrich ravens · Mind of the Raven · Ravens in Winter · raven cognition research

Do American crows use tools like New Caledonian crows?

Not at the same level. New Caledonian crows manufacture hook-shaped tools from leaf material and use sequences of tools to solve problems. American crows show occasional tool use in wild and captive settings but not at the species-typical level seen in New Caledonian crow. The species are both in the Corvus genus but show dramatically different cognitive specializations.

Read more inAuguste von Bayern and the New Caledonian crow tool programNew Caledonian crow tool use · Auguste von Bayern · Corvus moneduloides · crow tool making

Who is Auguste von Bayern?

A corvid cognition researcher at the Max Planck Institute for Ornithology in Germany. Her group, in collaboration with Alex Kacelnik's group at Oxford, has run a captive-population research program on New Caledonian crows for over a decade, producing many of the most-cited findings on the species's tool use and underlying cognitive capacities.

Read more inAuguste von Bayern and the New Caledonian crow tool programNew Caledonian crow tool use · Auguste von Bayern · Corvus moneduloides · crow tool making

Why are New Caledonian crows such good tool users?

A combination of ecological pressure (grub-rich tree crevices requiring extraction tools) and absence of competing tool-using species in their native habitat. The cognitive substrate for tool use exists across the Corvus genus; the New Caledonian species has had specific selection pressure that has elaborated the substrate into species-typical sophisticated tool behavior. Other Corvus species (including American crow) share less elaborated versions of the same cognitive machinery.

Read more inAuguste von Bayern and the New Caledonian crow tool programNew Caledonian crow tool use · Auguste von Bayern · Corvus moneduloides · crow tool making

Why is bird taxonomy still being revised?

DNA evidence allows direct measurement of genetic distance between populations, which often reveals patterns that morphology missed. The cumulative effect across the past three decades has been ongoing taxonomic revision — some species being split based on genetic evidence, others being merged. The species concept itself is in flux at the edge cases, contributing to the revisions.

Read more inBird taxonomy in flux: what counts as a speciesbird species taxonomy · crow species list · carrion crow hooded crow split · northwestern crow merger

Are carrion crow and hooded crow the same species?

Currently treated as separate species (Corvus corone and Corvus cornix) since around 2008-2015 across most authoritative taxonomies. They hybridize in a narrow zone in central Europe but are genetically distant enough that the biological species concept supports separation. The two were historically considered a single species with two color forms; the split reflects accumulated DNA evidence.

Read more inBird taxonomy in flux: what counts as a speciesbird species taxonomy · crow species list · carrion crow hooded crow split · northwestern crow merger

Is northwestern crow a separate species from American crow?

No, as of the 2020 American Ornithological Society revision. Northwestern crow has been merged into American crow as a subspecies (Corvus brachyrhynchos caurinus). Pacific Northwest crow populations are now considered the same species as the broader American crow, just a regional subspecies with slightly distinct morphology and vocalizations.

Read more inBird taxonomy in flux: what counts as a speciesbird species taxonomy · crow species list · carrion crow hooded crow split · northwestern crow merger

How do I identify individual crows?

Subtle plumage differences (slightly different bill shape, slight wing-feather molt patterns, occasional visible bands from researcher capture). Behavioral context (specific perch sites, specific times). Vocal characteristics develop in distinctive ways once you've heard the same individuals enough. Individual recognition is gradual: a few weeks of focused observation rather than instant identification. Many casual observers never get there; the atlas's approach assumes you can if you commit the time.

Read more inListening to your local crows: a practical guidelisten to crows · identify crow calls · neighborhood crows · learn crow vocalizations

Is it OK to record my local crows?

Yes, recording wild birds for personal observation, citizen science contribution, or documentation is generally appropriate. Avoid audio playback to attract birds (this is increasingly considered ethically inappropriate; see 'no playback for wild crows'). Don't approach nests or otherwise disturb the birds for the sake of recording. Respect distance and the birds' behavior — if they alarm or move away, you're too close.

Read more inListening to your local crows: a practical guidelisten to crows · identify crow calls · neighborhood crows · learn crow vocalizations

Where should I submit good recordings?

Wikimedia Commons for truly open Creative Commons licensing that supports projects like CrowLingo. eBird-Macaulay for population research contribution. iNaturalist for broader wildlife audio. Each platform has different licensing and usage models; Wikimedia Commons most directly supports open-data projects but is smaller scale than the Cornell-affiliated platforms.

Read more inListening to your local crows: a practical guidelisten to crows · identify crow calls · neighborhood crows · learn crow vocalizations

Why are crows associated with death in Western culture?

The association inherits primarily from European Christian and folk traditions. Some medieval European folklore associated crows with death omens and witchcraft. The association isn't ecological or behavioral fact; it's cultural inheritance. Other cultures (Pacific Northwest Indigenous, Norse, Japanese, others) have very different crow associations, including positive ones emphasizing cognition, memory, and guidance.

Read more inCrows in human cultures: a brief surveycrow symbolism cultures · crow mythology · Native American crow · Norse crow Hugin Munin

Do other cultures see crows differently?

Yes, substantially. Pacific Northwest Indigenous traditions feature Raven as a creator/trickster figure with cognitive complexity. Norse tradition pairs ravens with thought and memory. Japanese tradition includes the three-legged crow Yatagarasu as a divine messenger. The cultural variation in crow associations is wide; the 'bad omen' framing is one cultural inheritance, not a universal.

Read more inCrows in human cultures: a brief surveycrow symbolism cultures · crow mythology · Native American crow · Norse crow Hugin Munin

Should cultural meanings affect how I observe crows?

Ideally, no. Accurate observation works better when cultural pre-loading is suspended. Crows are highly social, cognitively complex, vocally rich generalist omnivores — that's what the species is, regardless of what cultural traditions have made of it. Recognizing your own cultural inheritance is part of doing accurate observation; the goal is observational attention without pre-loaded framing.

Read more inCrows in human cultures: a brief surveycrow symbolism cultures · crow mythology · Native American crow · Norse crow Hugin Munin

Why do crows roost together in such large numbers?

Multiple plausible functions: predator avoidance (safety in numbers), information sharing about food locations, possible thermoregulation benefits in cold weather, and social network maintenance across territories. The actual function probably combines multiple factors with different roosts emphasizing different ones based on local ecological conditions.

Read more inThe Bothell roost and other mega-aggregationsBothell crow roost · American crow communal roost · winter crow roost · why crows roost together

What's the Bothell crow roost?

A major American crow winter roost on the University of Washington Bothell campus outside Seattle. Peaks at approximately 15,000-20,000 individuals in winter months. Has been studied in detail by the Marzluff research group. The daily late-afternoon fly-in is visible to anyone on campus and is one of the more dramatic wildlife spectacles in the Pacific Northwest.

Read more inThe Bothell roost and other mega-aggregationsBothell crow roost · American crow communal roost · winter crow roost · why crows roost together

Are there other notable crow roosts to visit?

Yes. Auburn, Washington (King County Fairgrounds); Sacramento, California (downtown); Portland, Oregon (smaller but central); and many other Pacific Northwest sites. Smaller communal roosts exist across the species's range. The Pacific Northwest tends toward unusually-large roosts. Visiting in late afternoon during winter months offers the best viewing.

Read more inThe Bothell roost and other mega-aggregationsBothell crow roost · American crow communal roost · winter crow roost · why crows roost together

Do crows really play?

Yes, with methodological support beyond casual observation. Corvid behaviors including sliding down snowy surfaces, object manipulation with goal-directed structure, aerial aerobatics, and social play meet Gordon Burghardt's five formal criteria for play behavior. The play interpretation is not just anthropomorphism; it's an ethologically supported categorization.

Read more inCrow play behavior: what it shows about cognitioncrow play behavior · do crows play · corvid play · animal play cognition

What does play behavior show about crow intelligence?

Play behavior is correlated across animal species with cognitive complexity. The kinds of play corvids show — object play, social play, exploratory play — track the cognitive sophistication that other research lines (face recognition, episodic-like memory, social learning) establish for the species. Play is one of multiple converging lines of evidence for corvid cognitive complexity.

Read more inCrow play behavior: what it shows about cognitioncrow play behavior · do crows play · corvid play · animal play cognition

Is the snowboarding crow video real?

Yes, and similar behaviors had been previously reported in the academic literature for multiple corvid species. The viral video was a vivid documentation of behavior researchers were already aware of, not a new observation. Heinrich and Marzluff both describe similar play patterns in their books.

Read more inCrow play behavior: what it shows about cognitioncrow play behavior · do crows play · corvid play · animal play cognition

Do crows sleep all night?

Mostly yes, with occasional exceptions. Communal roost sites are largely quiet from full dark until pre-dawn. Exceptions include predator disturbance responses (especially to great horned owls), brief disturbance vocalizations during the night, and gradually-increasing pre-dawn vocalizations that build into the dawn-departure pattern. The species doesn't sustain through-the-night vocalization the way nocturnal species do.

Read more inCrow sleep and roost acousticscrow sleep · bird sleep behavior · roost acoustics · nocturnal crow vocalizations

What's the dawn departure?

The coordinated pattern by which crows depart a communal roost in the morning. Begins with quiet individual vocalizations in pre-dawn light, accelerates as dawn approaches, eventually includes group flight departures along consistent paths to daytime territories. Typically takes 30-60 minutes from first vocalization to substantial roost emptying. The acoustic record of this period is distinctive and amenable to AI analysis.

Read more inCrow sleep and roost acousticscrow sleep · bird sleep behavior · roost acoustics · nocturnal crow vocalizations

Could acoustic monitoring at roost sites add to research?

Yes, substantially. Continuous PAM deployment at American crow roosts would capture the full daily cycle (including under-observed night and pre-dawn periods) and generate longitudinal data on nocturnal disturbance, dawn-departure coordination, seasonal variation, and possible early indicators of population health. The methodology is mature; the research questions are open.

Read more inCrow sleep and roost acousticscrow sleep · bird sleep behavior · roost acoustics · nocturnal crow vocalizations

Have humans and crows always lived together?

Yes, at least throughout American crow's North American range over the past 10,000 years. Indigenous human populations occupied the continent for at least 15,000 years; American crow has existed throughout that period; the species was synanthropic (associated with human settlements) long before European colonization. The coevolution is older than usually appreciated.

Read more inCrows and humans: a coevolution sketchcrow human coevolution · synanthropic species history · American crow human history · commensal birds humans

Did crows benefit or suffer from European colonization?

Both. European agricultural and urban expansion produced landscapes that favored crow ecology. The species's population reached probably-unprecedented levels by the twentieth century. But hunting pressure was also substantial — bounty programs, organized shoots, federal pest-control authorization. The net was probably population increase across the twentieth century, followed by the 1999 West Nile virus impact, then partial recovery.

Read more inCrows and humans: a coevolution sketchcrow human coevolution · synanthropic species history · American crow human history · commensal birds humans

Why is American crow so wary of humans?

The species's learned-threat behavioral system, documented in detail by the Marzluff lab face-recognition work, is partially a product of long history with human hunting pressure. Crows learn to recognize specific threatening humans and treat them as targets for mobbing behavior; the capacity is species-typical, the specific threats are learned. The wariness is well-tuned to the actual historical context of crow-human relations.

Read more inCrows and humans: a coevolution sketchcrow human coevolution · synanthropic species history · American crow human history · commensal birds humans

Section

Journal · Methods

59 Q

How AI bioacoustics actually works: foundation models, embeddings, spectrograms, evaluation discipline — from the long-form journal.

Has AI translated crow language in 2026?

No. AI has mapped American crow vocalizations into a high-dimensional embedding space where similar calls cluster geometrically, but cluster identity is not lexical meaning. Translation, in the linguistic sense, would require synchronized behavioral evidence that crows treat sequence order as meaning-bearing, and that evidence remains thin. Be skeptical of any source claiming AI translates animal language today.

Read more inHow AI is decoding crow vocalizations in 2026AI crow vocalization · self-supervised audio · NatureLM-audio · Perch 2.0

Which AI model is best for crow vocalization analysis?

Perch 2.0 from Google Research is the current published state of the art for within-species detail in 2026. BirdNET remains dominant for real-time detection on consumer hardware. NatureLM-audio is best for natural-language Q&A about audio clips.

Read more inHow AI is decoding crow vocalizations in 2026AI crow vocalization · self-supervised audio · NatureLM-audio · Perch 2.0

What is a vocal map?

A vocal map is a 2D projection (typically UMAP) of a 1,024- or 1,536-dimensional audio embedding space. Each point is one vocalization; clusters are dense regions of acoustically similar calls. The geometry is discovered by the model, not assigned by humans.

Read more inHow AI is decoding crow vocalizations in 2026AI crow vocalization · self-supervised audio · NatureLM-audio · Perch 2.0

How many types of crow vocalizations are there?

Read more inHow AI is decoding crow vocalizations in 2026AI crow vocalization · self-supervised audio · NatureLM-audio · Perch 2.0

What is self-supervised learning in audio?

A training paradigm where a model learns from unlabeled audio by predicting masked-out portions of a spectrogram from the rest. The model develops rich learned representations as a side-effect of getting good at the prediction task. Used in BirdNET, Perch 2.0, and NatureLM-audio.

Read more inSelf-supervised audio learning, explained for non-engineersself-supervised learning · SSL audio · masked spectrogram prediction · audio foundation model

Why is self-supervised learning important for bioacoustics?

Labeled wildlife audio is scarce and expensive. Self-supervised models pretrain on the much larger pool of unlabeled wildlife recordings and only need labels for the final downstream task, breaking the labeled-set bottleneck that constrained pre-2020 bioacoustic AI.

Read more inSelf-supervised audio learning, explained for non-engineersself-supervised learning · SSL audio · masked spectrogram prediction · audio foundation model

Does self-supervised learning let AI understand crow language?

No. It lets AI produce rich acoustic embeddings — high-dimensional vectors capturing what makes calls similar or different. Interpreting those clusters in behavioral or semantic terms requires synchronized observation data, which is a separate scientific problem.

Read more inSelf-supervised audio learning, explained for non-engineersself-supervised learning · SSL audio · masked spectrogram prediction · audio foundation model

Which bioacoustic model should I use for crow research?

Perch 2.0 from Google Research is the current state of the art for within-species detail in 2026 — best for similarity search, clustering, individual ID, and behavioral classification. Use BirdNET for upstream detection and NatureLM-audio for natural-language questions about clips.

Read more inBirdNET vs Perch 2.0 vs NatureLM-audio: the practical 2026 guideBirdNET vs Perch · Perch 2.0 · NatureLM-audio · bioacoustic model comparison

Is BirdNET still relevant in 2026?

Yes. BirdNET remains dominant for real-time detection on consumer hardware, long-term acoustic monitoring deployments, and phone-based citizen science via Merlin. Its species coverage is the broadest of the three models.

Read more inBirdNET vs Perch 2.0 vs NatureLM-audio: the practical 2026 guideBirdNET vs Perch · Perch 2.0 · NatureLM-audio · bioacoustic model comparison

What is NatureLM-audio?

Earth Species Project's audio-language foundation model, ICLR 2025. BEATs audio encoder plus Llama-3.1-8B-Instruct language backbone. Answers natural-language questions about audio clips zero-shot. SOTA on BEANS-Zero.

Read more inBirdNET vs Perch 2.0 vs NatureLM-audio: the practical 2026 guideBirdNET vs Perch · Perch 2.0 · NatureLM-audio · bioacoustic model comparison

Can I run these models on a laptop?

BirdNET runs comfortably on a CPU laptop. Perch 2.0 needs a GPU for fast inference (an ONNX port helps for edge deployment). NatureLM-audio needs an A10G or A100; laptop inference is impractical.

Read more inBirdNET vs Perch 2.0 vs NatureLM-audio: the practical 2026 guideBirdNET vs Perch · Perch 2.0 · NatureLM-audio · bioacoustic model comparison

Why is animal-language AI harder than human-language AI?

Five structural differences: no parallel corpus, no native interpreter, no shared embodiment prior, no functional ground truth at the utterance level, and vastly less data. The first is the biggest — machine translation between human languages depends on millions of professionally-translated pairs, and animal communication has no analog.

Read more inWhy animal-language AI is harder than human-language AIanimal language AI · interspecies communication AI · bioacoustic deep learning limits · why animal AI is hard

Will AI translate animal language in five years?

No. The current bottleneck isn't algorithms; it's behavioral observation at scale that would let models learn receiver-side patterns. Wearable bioacoustic loggers are starting to produce that data for some species; even on optimistic timelines, translation in the linguistic sense remains far off.

Read more inWhy animal-language AI is harder than human-language AIanimal language AI · interspecies communication AI · bioacoustic deep learning limits · why animal AI is hard

What can animal-language AI actually do today?

Map vocal repertoires at fine resolution, identify individuals from short clips, characterize behavioral-context statistics at the cluster level, surface graded variation that hand-labeling missed. These are real capabilities — they're just not translation.

Read more inWhy animal-language AI is harder than human-language AIanimal language AI · interspecies communication AI · bioacoustic deep learning limits · why animal AI is hard

What is the Merlin app?

Merlin Sound ID is a free Cornell Lab of Ornithology phone app that identifies birds in real time by their vocalizations. Powered by BirdNET, it runs on modest phone hardware, works offline, and supports thousands of species across regions. Downloaded tens of millions of times globally as of 2025.

Read more inHow AI changed birdwatching: BirdNET, Merlin, and the citizen-science boomBirdNET Merlin app · AI birdwatching · Cornell bird ID · citizen science bioacoustics

Is Merlin the same as BirdNET?

Merlin Sound ID is built on BirdNET. BirdNET is the underlying detection model (developed by Stefan Kahl and colleagues at Cornell Lab and Chemnitz UT); Merlin is the user-facing app that packages BirdNET with a phone-friendly UX and eBird database integration.

Read more inHow AI changed birdwatching: BirdNET, Merlin, and the citizen-science boomBirdNET Merlin app · AI birdwatching · Cornell bird ID · citizen science bioacoustics

How accurate is Merlin?

Highly accurate for clean, near-field recordings of well-trained species in supported regions. Accuracy degrades with distance, ambient noise, overlapping calls, and underrepresented species. Used at scale for citizen-science presence/absence monitoring; not a substitute for research-grade acoustic analysis where individual ID, behavior, or graded variation matter.

Read more inHow AI changed birdwatching: BirdNET, Merlin, and the citizen-science boomBirdNET Merlin app · AI birdwatching · Cornell bird ID · citizen science bioacoustics

What do the axes on a UMAP plot mean?

Nothing specific. UMAP axes are abstract coordinates that the algorithm chose to position dots in a way that preserves their high-dimensional relationships. They don't represent named acoustic properties like pitch or duration. The information is in the relationships between dots, not in absolute positions.

Read more inReading the vocal atlas: a beginner's guide to UMAP for bird soundsUMAP bird sounds · vocal atlas reading · audio embedding visualization · how to read UMAP

Why are there clusters in the vocal atlas?

Dense regions of the 2D scatter correspond to acoustically similar groups of recordings. For crows, those clusters mostly correspond to call types: territorial, mobbing, assembly, rattle, begging, etc. The cluster names are assigned by humans matching exemplars against the descriptive literature; the model knows the geometry, humans assign the names.

Read more inReading the vocal atlas: a beginner's guide to UMAP for bird soundsUMAP bird sounds · vocal atlas reading · audio embedding visualization · how to read UMAP

Is the UMAP scatter the same as the full embedding?

No. UMAP is a 2D projection of a 1,024-or-1,536-dim embedding space. It preserves local neighborhood structure but loses some global geometry. For visualization, the projection is useful; for similarity search and nearest-neighbor queries, work with the full high-dimensional vectors.

Read more inReading the vocal atlas: a beginner's guide to UMAP for bird soundsUMAP bird sounds · vocal atlas reading · audio embedding visualization · how to read UMAP

What is a spectrogram?

A two-dimensional picture of sound. Time runs across the horizontal axis, frequency runs up the vertical axis, and brightness or color shows how much acoustic energy is present at each time-frequency point. The standard visualization in bioacoustic and acoustic research.

Read more inSpectrograms decoded: what those squiggles actually showhow to read a spectrogram · spectrogram explained · audio frequency time · FFT bird sound

Why do AI models use spectrograms instead of audio waveforms?

Because spectrograms preserve all the time-frequency information needed to distinguish sounds, in a 2D image-like format that image-recognition architectures can process directly. The convergence has been productive: bioacoustics inherited two decades of computer-vision research the moment it switched to spectrogram representation.

Read more inSpectrograms decoded: what those squiggles actually showhow to read a spectrogram · spectrogram explained · audio frequency time · FFT bird sound

What are harmonics on a spectrogram?

Horizontal stripes stacked above a fundamental frequency. Tonal sounds (like crow caws) produce a fundamental at one frequency plus integer multiples of it. The fundamental is the bottom stripe; harmonics are the higher stripes. Relative loudness of different harmonics — harmonic emphasis — is a key feature for individual identification in crow research.

Read more inSpectrograms decoded: what those squiggles actually showhow to read a spectrogram · spectrogram explained · audio frequency time · FFT bird sound

What is Voxaboxen?

An open-source Python library for bioacoustic vocalization segmentation, released by Earth Species Project. Given a long audio recording, it identifies the start and end times of individual vocalizations in it. Used in multiple recent corvid research papers including Demartsev et al. 2026.

Read more inVoxaboxen and the open infrastructure of modern bioacousticsVoxaboxen Earth Species · bioacoustic segmentation · ESP open source tools · audio annotation tool

Does Voxaboxen classify bird species?

No. Voxaboxen handles temporal segmentation only. Species and call-type classification is done by downstream audio foundation models like BirdNET, Perch 2.0, or NatureLM-audio. The pipeline is modular: segment first, then classify or embed.

Read more inVoxaboxen and the open infrastructure of modern bioacousticsVoxaboxen Earth Species · bioacoustic segmentation · ESP open source tools · audio annotation tool

Why is open-source infrastructure important for bioacoustics?

Because data acquisition is the bottleneck in bioacoustic research. Shared tooling lets each lab build on the same processing infrastructure rather than reinventing it, which means more researchers can attempt more ambitious questions and methodological refinements accumulate faster across the field.

Read more inVoxaboxen and the open infrastructure of modern bioacousticsVoxaboxen Earth Species · bioacoustic segmentation · ESP open source tools · audio annotation tool

What's the biggest misconception about AI animal communication research?

That mapping a vocal repertoire is the same as translating it. AI methods can map vocalizations geometrically and characterize cluster-level behavioral associations; they cannot translate without receiver-side behavioral evidence that the contemporary methods alone cannot produce.

Read more inWhat corvid researchers wish you understood about AI bioacousticscorvid researcher perspective · AI bioacoustics critique · what scientists say about animal AI · bioacoustics insider view

Are working scientists optimistic about translating animal language?

Not in the strong sense the popular framing implies. Most are optimistic about better repertoire mapping, more rigorous behavioral validation, and meaningful insights into communication structure. They are not generally optimistic about translation in the next decade with current methods.

What deployments matter most for AI bioacoustics?

Counter-intuitively, BirdNET-powered Merlin on phones (tens of millions of users) and AudioMoth-based passive acoustic monitoring (tens of thousands of devices) — both consumer/deployment scales — have moved the field more than any single research-grade foundation model release. Reaching the public reshapes the data infrastructure the research depends on.

What information do spectrograms lose?

Phase information (timing and shape of the underlying waveform), some fine-grained timbre detail, and resolution at the time-frequency trade-off. Different windowing choices produce different time/frequency resolution compromises; no single spectrogram captures the full underlying audio.

Read more inWhat spectrograms can't tell youspectrogram limitations · what spectrograms hide · FFT window resolution · spectrogram artifacts

Why do some spectrograms look so different from others of the same sound?

Window length (time vs. frequency resolution trade-off), frequency-axis scaling (linear vs. mel vs. log), color scheme dynamic range, and dB normalization all affect appearance. The same audio file can produce visually different spectrograms depending on these choices. Comparing spectrograms across sources requires noting these parameters.

Read more inWhat spectrograms can't tell youspectrogram limitations · what spectrograms hide · FFT window resolution · spectrogram artifacts

Do AI bioacoustic models use spectrograms or raw audio?

Most current models use mel-scaled spectrograms as input (BirdNET, Perch 2.0, BEATs lineage). A few experimental architectures operate on raw waveforms directly, which can potentially preserve phase information lost in spectrograms — but the spectrogram-based architectures dominate practical deployment.

Read more inWhat spectrograms can't tell youspectrogram limitations · what spectrograms hide · FFT window resolution · spectrogram artifacts

What is the Macaulay Library?

The world's largest scientific archive of wildlife audio, video, and photographs, housed at the Cornell Lab of Ornithology. Approximately 1.3 million recordings as of 2026, with growth driven by integration with the eBird citizen-science platform since the late 2000s.

Read more inThe Macaulay Library and the open-research tensionMacaulay Library Cornell · bioacoustics open data · ornithology archive · Cornell Lab data licensing

Is Macaulay Library data open?

Partially. Individual recordings are accessible for non-commercial individual use with attribution. Bulk downloads, programmatic access, and research-scale dataset use require Cornell-administered agreements. The licensing is not Creative Commons in the openly-redistributable sense — recordings are stewarded by Cornell, not released to the public domain.

Read more inThe Macaulay Library and the open-research tensionMacaulay Library Cornell · bioacoustics open data · ornithology archive · Cornell Lab data licensing

Why don't open AI models use Macaulay data directly?

Most foundation models (BirdNET, Perch, NatureLM-audio) train on Macaulay-derived data under research agreements but can't redistribute the underlying audio. End users can use the models without the training audio. CrowLingo specifically uses Wikimedia Commons CC-licensed audio to avoid this constraint and ensure the corpus is truly open.

Read more inThe Macaulay Library and the open-research tensionMacaulay Library Cornell · bioacoustics open data · ornithology archive · Cornell Lab data licensing

What is eBird?

A citizen-science platform launched by the Cornell Lab of Ornithology in 2002. Users submit checklists of bird observations from specific locations; the data aggregates into population-level analyses, range maps, and conservation tracking. eBird is the largest bird-observation database in the world, with over a billion observations to date.

Read more inHow eBird became a bioacoustic engineeBird audio recordings · eBird bioacoustics · citizen science bird audio · Cornell Lab eBird

How did eBird become a major source of bioacoustic data?

eBird added audio submission capability in the early 2010s; users could attach phone-recorded clips to checklists as documentation. The audio flowed automatically into the Macaulay Library. Once mobile phone recording became universal around 2015, eBird became the largest active feeder of bioacoustic data globally, providing most of the recent additions to Macaulay's 1.3-million-recording archive.

Read more inHow eBird became a bioacoustic engineeBird audio recordings · eBird bioacoustics · citizen science bird audio · Cornell Lab eBird

Are eBird audio recordings open data?

They carry Macaulay Library licensing — generally available for individual non-commercial use with attribution, but not Creative Commons in the openly-redistributable sense. Bulk download and large-scale research use require Cornell-administered agreements. This creates a useful but partially-proprietary data infrastructure for the bioacoustics field.

Read more inHow eBird became a bioacoustic engineeBird audio recordings · eBird bioacoustics · citizen science bird audio · Cornell Lab eBird

What is passive acoustic monitoring?

Deploying autonomous recording devices in the field to capture continuous or scheduled audio from a study area, with subsequent offline analysis by classification models. The hardware (often AudioMoth devices at $80-100 per unit) runs on batteries for weeks; the analysis identifies species presence and abundance using models like BirdNET. PAM has matured into a standard conservation tool over the past decade.

Read more inAcoustic monitoring as a conservation toolpassive acoustic monitoring · PAM conservation · soundscape ecology · AudioMoth conservation

Why is PAM changing conservation biology?

It allows monitoring wildlife communities at much larger spatial and temporal scales than human-observer surveys. A small lab can deploy dozens of recorders covering sites observers rarely visit, recording continuously for weeks. The combination of cheap hardware (AudioMoth and similar) and capable AI classification (BirdNET, Perch 2.0) makes large-scale, fine-grained, multi-species inventories practical in ways that were impossible just ten years ago.

Read more inAcoustic monitoring as a conservation toolpassive acoustic monitoring · PAM conservation · soundscape ecology · AudioMoth conservation

What are the limitations of acoustic monitoring?

Detection probability varies by species (quiet species systematically under-detected). Abundance estimation requires careful modeling of variable detection probabilities. Species identification by AI is good but not perfect. Acoustic monitoring measures vocalizing individuals, missing non-vocalizing ones. These constraints are real but manageable; the methodology produces useful science within them.

Read more inAcoustic monitoring as a conservation toolpassive acoustic monitoring · PAM conservation · soundscape ecology · AudioMoth conservation

What is a syrinx?

The sound-producing organ unique to birds, located at the bottom of the trachea where the airway splits into the two bronchi. Membranes vibrate as air flows past them; surrounding muscles control airflow and membrane tension. The syrinx is anatomically distinct from the mammalian larynx (which is at the top of the trachea), and the structural differences shape what bird vocalizations can do.

Read more inHow birds actually produce sound: the syrinx briefly explainedbird syrinx · how do birds sing · avian sound production · Coen Elemans syrinx

Can birds really sing two notes at once?

Yes, in many songbird species. The syrinx has two sides (one on each bronchus) that can be controlled semi-independently. Wood thrushes, song sparrows, and many other songbirds use this capability for the structural complexity of their songs. American crow uses the two-voice capability less than typical singing songbirds, biasing the species toward shorter impulsive calls rather than sustained two-voice songs.

Read more inHow birds actually produce sound: the syrinx briefly explainedbird syrinx · how do birds sing · avian sound production · Coen Elemans syrinx

Why does syrinx anatomy matter for AI bioacoustic models?

Bird vocalizations carry signatures of the physiological system that produced them — frequency range, call-shape patterns, timing precision, spectral richness are all constrained by syringeal anatomy. AI models learn these patterns implicitly. Understanding the underlying biology helps researchers know when AI models are finding real biological structure versus acoustic artifacts, and helps with cross-species generalization.

Read more inHow birds actually produce sound: the syrinx briefly explainedbird syrinx · how do birds sing · avian sound production · Coen Elemans syrinx

When did wildlife audio recording start?

Albert R. Brand founded the Macaulay Library at Cornell in 1929, marking the start of systematic wildlife audio collection. Early recordings used phonograph cylinders and discs. Magnetic tape became dominant in the 1950s; digital recording in the 1990s and 2000s; mobile-phone citizen-science recordings in the 2010s; autonomous PAM recorders from the late 2010s onward.

Read more inThe history of bird recording technologybird recording history · wildlife audio technology · Albert Brand Cornell · bioacoustics equipment evolution

What is an AudioMoth?

An open-source acoustic recorder developed by Open Acoustic Devices, released in 2016. Costs $80-100 per unit. Runs on AA batteries for weeks. Programmable for scheduled or trigger-based recording. Has become the iconic example of cheap autonomous wildlife recording, enabling spatial-and-temporal-scale acoustic monitoring that wasn't practical before its release.

Read more inThe history of bird recording technologybird recording history · wildlife audio technology · Albert Brand Cornell · bioacoustics equipment evolution

Are wildlife audio archives unbiased?

No, and recognizing the biases is part of using the data well. Under-representation of nocturnal species, quiet species, and remote locations. Geographic bias toward English-speaking-country contributors and populated regions. Quality variation reflecting the multiple eras of recording technology that contributed to the archives. AI models trained on this data inherit these biases; the bias is a normal feature of citizen-science-augmented data infrastructure, not a flaw to be hidden.

Read more inThe history of bird recording technologybird recording history · wildlife audio technology · Albert Brand Cornell · bioacoustics equipment evolution

Do I need professional recording equipment to contribute citizen-science audio?

No. Phone recordings are acceptable on most citizen-science platforms (eBird, iNaturalist, Wikimedia Commons). The quality bar is more about clarity of the target signal than about technical recording quality. A clear phone recording of a crow alarm call from 20 feet away is more useful than a high-end field recording with poor signal isolation.

Read more inWhat good bioacoustic citizen science looks likebioacoustic citizen science · eBird recording best practices · wildlife audio submission · Macaulay Library contribution

What makes a citizen-science recording valuable?

Clear primary signal, reasonable duration (10-60 seconds typically), accurate location and time metadata, species identification confidence (or honest flagging when uncertain), continuous recording rather than just snippets, behavioral context notes where observable, and contribution at locations and times that the existing archive under-represents.

Read more inWhat good bioacoustic citizen science looks likebioacoustic citizen science · eBird recording best practices · wildlife audio submission · Macaulay Library contribution

Where should I submit recordings?

eBird-Macaulay for general bird audio that supports North American population research. iNaturalist for broader wildlife audio across taxa. Wikimedia Commons for recordings you want under truly-open Creative Commons licensing for unrestricted downstream use. CrowLingo's atlas specifically uses Wikimedia Commons, so adding open-license recordings there directly supports the open bioacoustic data ecosystem.

Read more inWhat good bioacoustic citizen science looks likebioacoustic citizen science · eBird recording best practices · wildlife audio submission · Macaulay Library contribution

Can I understand bioacoustics papers without specialist training?

Mostly yes, with appropriate orientation. The structure is consistent enough that non-specialists can read papers productively: abstract for the claim, methods for the kind of evidence, results for the findings, discussion for interpretation, limitations for the caveats. The goal isn't to verify every statistical detail; it's to understand what claim is being made, what evidence supports it, and how confident the result is.

Read more inHow to read a bioacoustics paperhow to read bioacoustics paper · scientific paper reading guide · bioacoustic methodology · research paper layperson guide

What are the three rough types of bioacoustics papers?

Descriptive ('we documented behavior X in species Y'), experimental ('we tested hypothesis Z and found evidence supporting/contradicting it'), and methodological ('we developed method M and demonstrated it on dataset D'). Identifying which type a paper is helps orient your reading: descriptive papers establish what exists; experimental papers test causal claims; methodological papers establish how subsequent research can proceed.

Read more inHow to read a bioacoustics paperhow to read bioacoustics paper · scientific paper reading guide · bioacoustic methodology · research paper layperson guide

Why is the limitations section so important?

It's where authors tell you, in their own words, what the study can't conclude. Most rigorous bioacoustics papers include explicit limitations sections to pre-empt peer-reviewer objections. Reading the limitations carefully tells you what the work actually establishes versus what would require additional research. A paper without a limitations section or with only cursory acknowledgment is a flag for potential over-claiming.

Read more inHow to read a bioacoustics paperhow to read bioacoustics paper · scientific paper reading guide · bioacoustic methodology · research paper layperson guide

Are BirdNET, Perch, and NatureLM-audio really open source?

Partially, with variation. All three are open on model weights and architecture. Training code is partially documented for all three with varying completeness. Training data is the most-constrained layer across all three because it includes Macaulay Library and similar archives with licensing constraints. The 'open source' framing is accurate for the weights-and-architecture layers and partially accurate for the code and data layers.

Read more inWhat 'open source' actually means in bioacoustic AIopen source bioacoustic AI · BirdNET open source · Perch open source · NatureLM-audio open source

Can I use these models commercially?

Generally yes, depending on the specific license of the model weights. BirdNET uses Apache 2.0 which is commercial-friendly. Perch 2.0 and NatureLM-audio have their own licenses that should be checked individually for specific use cases. The model weights' license is the critical layer for commercial use; downstream model use generally doesn't require interaction with the more-constrained training-data layer.

Read more inWhat 'open source' actually means in bioacoustic AIopen source bioacoustic AI · BirdNET open source · Perch open source · NatureLM-audio open source

Why does training-data openness matter?

Several reasons. Reproducibility: fully reproducing a model from scratch requires the training data. Auditability: checking the model for biases or coverage gaps requires inspecting what it learned from. Downstream open-source projects: building fully-open downstream work depends on training data being available for redistribution. Most current bioacoustic foundation models have constraints at the training-data layer that complicate these uses.

Read more inWhat 'open source' actually means in bioacoustic AIopen source bioacoustic AI · BirdNET open source · Perch open source · NatureLM-audio open source

What is Stowell 2022?

Dan Stowell's review paper 'Computational bioacoustics with deep learning: a review and roadmap,' published in PeerJ Computer Science in 2022. The most-cited methodological reference in modern bioacoustic AI work. Synthesized the state of the field, identified methodological priorities, and helped set the agenda for subsequent work. Open-access and freely available.

Read more inStowell 2022: the deep-learning bioacoustics review that grounds the fieldStowell 2022 review · deep learning bioacoustics · PeerJ bioacoustic review · audio classification AI review

What did the review argue?

Three main claims. Deep learning had transformed computational bioacoustics over the prior five years and was decisively outperforming classical signal-processing approaches. The deep-learning architectures that worked best converged on a small set of patterns (CNN backbones, transformers, attention mechanisms). Methodological challenges remained, particularly around training-data biases, evaluation protocols, and the gap between benchmark and field-deployment performance.

Read more inStowell 2022: the deep-learning bioacoustics review that grounds the fieldStowell 2022 review · deep learning bioacoustics · PeerJ bioacoustic review · audio classification AI review

Is this paper still relevant in 2026?

Yes, as the foundational synthesis reference. Some specific methodology has advanced (Perch 2.0 cross-domain work, NatureLM-audio self-supervised pretraining), but the basic framework Stowell laid out remains current. For technically-oriented readers wanting to understand how bioacoustic AI works, the review is the recommended entry point. Most subsequent papers in the field cite it as their methodological grounding.

Read more inStowell 2022: the deep-learning bioacoustics review that grounds the fieldStowell 2022 review · deep learning bioacoustics · PeerJ bioacoustic review · audio classification AI review

Section

Journal · Decoding

15 Q

What can be seen in crow vocalizations now with AI audio embeddings, and what can't — from the long-form journal.

Do American crows have dialects?

Yes, in the descriptive sense. Family groups carry measurable acoustic signatures that differ between geographically-separated populations. Whether those differences function as cultural dialect — shaping how crows respond to one another — is not yet proven.

Read more inDo crows have dialects? What the evidence actually showscrow dialects · American crow communication · bird dialect evidence · vocal culture corvids

Can AI identify individual crows by voice?

Yes. Harmonic emphasis — the relative loudness of second and third harmonics versus the fundamental — fingerprints individuals with accuracy that approaches the limit set by recording quality, not by audio content.

Read more inDo crows have dialects? What the evidence actually showscrow dialects · American crow communication · bird dialect evidence · vocal culture corvids

What's the best evidence that crow dialects are cultural?

Multi-year studies showing that inter-group acoustic differences exceed within-group variation, correlate with geographic distance, and aren't explained by population genetic structure. The remaining uncertainty is whether crows themselves use those differences functionally.

Read more inDo crows have dialects? What the evidence actually showscrow dialects · American crow communication · bird dialect evidence · vocal culture corvids

What is the Mates 2014 paper?

Eva Mates, Tarter, Ha, Clark, and McGowan's 2014 Bioacoustics paper showing American crow caws acoustically encode caller individual identity, with harmonic emphasis (the relative loudness of second and third harmonics versus the fundamental) as the most reliable individual discriminator. The canonical reference for modern AI work on crow identity.

Read more inThe Mates 2014 paper and why it still mattersMates 2014 · crow caw individual identity · American crow acoustic profiling · McGowan Cornell crows

Why is Mates 2014 still cited if the methods are dated?

Because the finding is method-independent. Modern embedding pipelines reproduce Mates et al.'s individual-identity result with newer methods; the empirical anchor matters more than the original method. New AI methods need anchors to distinguish real findings from method artifacts.

Read more inThe Mates 2014 paper and why it still mattersMates 2014 · crow caw individual identity · American crow acoustic profiling · McGowan Cornell crows

Did Mates et al. show crows have dialects?

No. The 2014 paper showed individual identity is acoustically recoverable. Subsequent work has extended toward dialect at the group level (the layer-two question), but Mates et al. themselves stayed at the layer-one individual-identity finding, which is the harder claim to defend and the one that holds best across methodological changes.

Read more inThe Mates 2014 paper and why it still mattersMates 2014 · crow caw individual identity · American crow acoustic profiling · McGowan Cornell crows

What can AI actually decode about American crow communication today?

Caller sex (high confidence), individual identity (high confidence), behavioral context at the cluster level (medium-high confidence), family-group dialect at the descriptive level (medium confidence). AI cannot decode lexical meaning, translate calls into language, or definitively establish compositional structure with the contemporary methods.

Read more inWhat we can and cannot decode about crow communication, todaywhat AI can decode about crows · crow communication AI 2026 · limits of bioacoustics AI · scientific calibration animal language

What can AI not decode about crow communication?

Lexical meaning of specific calls, translation in any linguistic sense, definitive evidence of compositional structure, functional dialect (whether crows themselves use group-level acoustic differences). These claims would require receiver-side behavioral evidence the contemporary AI methods cannot produce alone.

Why does popular coverage keep over-promising?

Because the headline is irresistible and the science doesn't fit neatly into popular framings. Most over-promising articles conflate cluster-level statistical associations (which AI does well) with utterance-level meaning translation (which AI cannot do).

What is the biggest open question in crow vocal research?

The receiver-side decoding problem — whether crows, hearing a specific call, make specific inferences that demonstrably change their behavior. We have strong sender-side characterization (calls carry individual identity, cluster by behavioral context). We have weak receiver-side evidence. The gap is the largest unsolved question and the focus of the next decade of meaningful work.

Read more inWhat we don't know about crow communicationunanswered questions corvid · what we don't know crows · open questions bioacoustics · crow research gaps

Do AI methods reveal everything about crow vocalizations?

No. AI embedding methods optimize for what their training distribution rewards (mostly species discrimination and broad acoustic similarity). Whether they miss dimensions of variation that classical methods captured is an open question. Most of the field assumes the answer is 'unlikely to be significant'; the assumption hasn't been rigorously stress-tested.

Read more inWhat we don't know about crow communicationunanswered questions corvid · what we don't know crows · open questions bioacoustics · crow research gaps

Why don't we know more about quiet crow vocalizations?

Most public audio corpora capture what humans can hear from twenty meters, which excludes the quiet grunts crows produce at close range. Wearable bioacoustic loggers (Demartsev 2026 for carrion crows) are starting to recover quiet vocalizations at scale; American crow studies are catching up.

Read more inWhat we don't know about crow communicationunanswered questions corvid · what we don't know crows · open questions bioacoustics · crow research gaps

Do crows have predator-specific alarm calls?

On the sender side, yes — measurable acoustic differences in mobbing calls correlate with predator type. On the receiver side, less clearly established — whether crows treat these acoustic differences as specific predator-type information remains debated. The question is harder to settle cleanly than for vervet monkeys because of methodological complications around visual context and individual variation.

Read more inPredator-specific alarm calls: do crows have words for hawks?predator specific alarm call · crow alarm call types · referential signals birds · vervet monkey alarm crows

What's the vervet monkey alarm call finding?

Robert Seyfarth and Dorothy Cheney's 1980 paper showing vervet monkeys produce distinct alarm calls for leopards, eagles, and snakes — calls that elicit predator-type-appropriate responses in conspecifics even when no predator is visible. The canonical example of referential animal communication.

Read more inPredator-specific alarm calls: do crows have words for hawks?predator specific alarm call · crow alarm call types · referential signals birds · vervet monkey alarm crows

Are crow alarm calls referential?

Sender-side variation suggests acoustic structure carries predator-type information; receiver-side decoding evidence is less clear. The question remains open and is one of the most-studied unresolved problems in corvid vocal research. The stakes are high because referential alarms are the closest animal-communication analogue to lexical meaning in human language.

Read more inPredator-specific alarm calls: do crows have words for hawks?predator specific alarm call · crow alarm call types · referential signals birds · vervet monkey alarm crows

Section

Journal · Pipeline

7 Q

From phone recording to interpretable vocal map — the bioacoustic AI pipeline explained step by step in the long-form journal.

What are the stages of a bioacoustic AI pipeline?

Eight: capture (microphone in front of the bird), detect (find the calls in a long recording), preprocess (filter, normalize, chunk), embed (audio foundation model produces a 1,024-or-1,536-dim vector), project (UMAP to 2D for inspection), cluster (HDBSCAN on full embeddings), contextualize (join to behavioral data), respond (decide what to do with the result, ethically constrained).

Read more inFrom caw to cluster: the eight-stage pipeline explainedbioacoustic pipeline · from caw to cluster · crow audio analysis pipeline · AI bioacoustics workflow

Which model should I use for the embed stage?

Perch 2.0 from Google Research is the current state of the art for within-species detail. BirdNET is faster for detection. NatureLM-audio is best for natural-language Q&A about clips. Don't mix encoders in the same UMAP — embedding spaces aren't interoperable.

Read more inFrom caw to cluster: the eight-stage pipeline explainedbioacoustic pipeline · from caw to cluster · crow audio analysis pipeline · AI bioacoustics workflow

Why HDBSCAN instead of K-means?

HDBSCAN doesn't require you to pick the cluster count in advance, allows variable-density clusters, and treats outliers as noise rather than forcing them into a category. For natural acoustic repertoires where cluster count is unknown and cluster density varies, HDBSCAN is the right primitive.

Read more inFrom caw to cluster: the eight-stage pipeline explainedbioacoustic pipeline · from caw to cluster · crow audio analysis pipeline · AI bioacoustics workflow

What microphone should I use to record crows?

A directional shotgun like the Sennheiser ME66 or MKE600, or a Rode NTG-2, with a foam-and-fur windscreen. Off-axis rejection matters more than absolute sensitivity.

Read more inHow to record crows: a field guide for citizen scientistshow to record crows · field recording crow audio · bioacoustics citizen science · Sennheiser ME66 crow

What sample rate and bit depth should I record at?

48 kilohertz, 24-bit, mono. Captures the full crow vocal range with headroom for any downstream filter. Mono is sufficient — most embedding models embed mono anyway.

Read more inHow to record crows: a field guide for citizen scientistshow to record crows · field recording crow audio · bioacoustics citizen science · Sennheiser ME66 crow

How close can I get to a crow nest?

No closer than ten meters from an active nest at any time of year. During breeding-season hours for the first three weeks after a nest is active, give even more distance. The disturbance literature is clear: 'just a few meters closer' is not benign.

Read more inHow to record crows: a field guide for citizen scientistshow to record crows · field recording crow audio · bioacoustics citizen science · Sennheiser ME66 crow

Where can I contribute my crow recordings?

Xeno-canto, Wikimedia Commons, or directly to research labs that accept contributions. CrowLingo accepts CC-BY-SA 4.0 contributions via contact@kymatalabs.com.

Read more inHow to record crows: a field guide for citizen scientistshow to record crows · field recording crow audio · bioacoustics citizen science · Sennheiser ME66 crow

Section

Journal · Frontier

45 Q

Where the field is going: Project CETI, NatureLM-audio, vocal learning, animal-language translation limits — from the long-form journal.

What is the Demartsev 2026 paper?

A bioRxiv preprint by Vlad Demartsev, Ariana Strandburg-Peshkin and colleagues mapping the carrion crow vocal repertoire using wearable bioacoustic loggers. 127,000+ vocalizations across behavioral contexts, attributed to specific known individuals, with synchronized behavioral observation. The cleanest contemporary example of where wild corvid bioacoustics is going.

Read more inWearable bioacoustic loggers and the Demartsev revolutionwearable bioacoustic logger · Demartsev 2026 · carrion crow research · synchronized behavioral audio

Does the Demartsev study apply to American crows?

Partially. American crows and carrion crows are sister species with similar social systems; findings about contextual modulation of call meaning should transfer in spirit. The methodology — wearable loggers + synchronized observation — generalizes directly. No equivalent American crow study has been published as of mid-2026.

Read more inWearable bioacoustic loggers and the Demartsev revolutionwearable bioacoustic logger · Demartsev 2026 · carrion crow research · synchronized behavioral audio

What is a wearable bioacoustic logger?

A small audio-recording device small enough to be carried by a wild bird without altering behavior, with onboard storage and a synchronized clock. Records the bird's own vocalizations at close range, eliminating ambient noise and per-individual attribution problems that constrained classical stationary-microphone studies.

Read more inWearable bioacoustic loggers and the Demartsev revolutionwearable bioacoustic logger · Demartsev 2026 · carrion crow research · synchronized behavioral audio

Can AI translate animal language in 2026?

No. Contemporary AI methods produce rich embeddings of animal vocalizations and can map vocal repertoires geometrically, but translation in the linguistic sense requires evidence that signals encode meaning compositionally and that receivers decode them — neither of which the embedding methods alone can establish.

Read more inWhat 'translating' animal language would actually requiretranslate animal language · AI animal communication · Project CETI sperm whale · Earth Species Project

What would be required to translate crow language?

Sender-side mapping (call features reliably accompany contexts), receiver-side mapping (crows treat features as carrying that information), reproducibility across populations and individuals, and a target representation that preserves meaning. No animal communication system has been credibly translated in this strong sense; honeybee waggle dance is the closest case, and it corresponds to spatial coordinates rather than language.

Read more inWhat 'translating' animal language would actually requiretranslate animal language · AI animal communication · Project CETI sperm whale · Earth Species Project

Why do popular articles keep saying AI will translate animal language?

Because the headline is irresistible and the science is undertested in the directions that would matter. Most popular pieces conflate vocal-repertoire mapping (which is real and improving rapidly) with translation (which has receiver-side requirements the mapping methods cannot satisfy).

Read more inWhat 'translating' animal language would actually requiretranslate animal language · AI animal communication · Project CETI sperm whale · Earth Species Project

What is Project CETI?

The Cetacean Translation Initiative — a multi-year, multi-million-dollar research project applying contemporary AI methods to sperm whale coda vocalizations. Led by Shafi Goldwasser, David Gruber, Jacob Andreas and colleagues across MIT, Harvard, the National Geographic Society, and the Dominica Sperm Whale Project.

Read more inProject CETI and what we learned from sperm whalesProject CETI sperm whale · cetacean translation initiative · Andreas 2022 CETI · sperm whale codas

Has CETI translated sperm whale language?

No, and the team has been explicit that it remains an open question whether translation is the right frame. The 2022 Andreas et al. roadmap paper named the epistemological risks — translation may not apply, meaning may not be lexical, structure may be absent where assumed — and the team's communication has tracked that posture.

Read more inProject CETI and what we learned from sperm whalesProject CETI sperm whale · cetacean translation initiative · Andreas 2022 CETI · sperm whale codas

What can CrowLingo learn from CETI?

Both projects face the same fundamental constraint: receiver-side meaning isn't observable from sender-side audio alone. Both have chosen careful cataloging over translation claims. The shared lesson is to map first and treat meaning as a separate, slower scientific problem.

Read more inProject CETI and what we learned from sperm whalesProject CETI sperm whale · cetacean translation initiative · Andreas 2022 CETI · sperm whale codas

What is passive acoustic monitoring (PAM)?

A methodology in which continuously-recording acoustic devices are deployed in a habitat to capture species vocalizations without active intervention. The recordings are analyzed automatically — typically with BirdNET or related models — to track species presence, abundance, phenology, or community structure over time.

Read more inPassive acoustic monitoring: what AI ears reveal about wild populationspassive acoustic monitoring · PAM bird conservation · AudioMoth deployment · long-term bioacoustic survey

What devices are used for passive acoustic monitoring?

AudioMoth (the open-source logger from Open Acoustic Devices) is the most common, shipping for under one hundred dollars per unit. Multiple commercial successors (Wildlife Acoustics SongMeter, Frontier Labs, others) extend the market for higher-spec deployments. Together they've dropped the per-unit cost of acoustic monitoring by more than an order of magnitude since 2015.

Read more inPassive acoustic monitoring: what AI ears reveal about wild populationspassive acoustic monitoring · PAM bird conservation · AudioMoth deployment · long-term bioacoustic survey

What can AI passive acoustic monitoring not do?

Reliable individual identification at scale, per-utterance behavioral context inference, accurate counts of individuals, replacement of ground-truth field surveys for population estimates. PAM is best at presence/absence and phenology questions; it's not a substitute for focal observation or wearable-logger work.

Read more inPassive acoustic monitoring: what AI ears reveal about wild populationspassive acoustic monitoring · PAM bird conservation · AudioMoth deployment · long-term bioacoustic survey

What is an acoustic index?

A mathematical summary of an audio recording designed to capture some aspect of the underlying ecosystem. Examples include the acoustic complexity index (ACI), acoustic diversity index (ADI), and bioacoustic index. Each reduces hours of audio to a single number or small vector for cross-site or cross-time comparison.

Read more inAcoustic indices: measuring ecosystem health through soundacoustic indices ecology · soundscape ecology · Bryan Pijanowski · acoustic complexity index

Do acoustic indices actually work?

Yes for coarse comparisons (healthy versus degraded habitats, urban versus wild sites, multi-year trend tracking). Less reliably for fine-grained questions about specific species or biodiversity composition. The cleanest published critiques caution against single-index summaries for high-stakes conservation decisions without ground-truth validation.

Read more inAcoustic indices: measuring ecosystem health through soundacoustic indices ecology · soundscape ecology · Bryan Pijanowski · acoustic complexity index

Are AI methods replacing acoustic indices?

Partially. Foundation-model embeddings averaged over a recording capture more compositional information than classical indices, but they're less interpretable. The field is converging on hybrid practice: foundation-model embeddings for compositional questions, classical indices for trend tracking.

Read more inAcoustic indices: measuring ecosystem health through soundacoustic indices ecology · soundscape ecology · Bryan Pijanowski · acoustic complexity index

Are American crow populations declining?

Not nationally, as of the early 2020s. Continent-wide numbers are roughly stable. Substantial regional variation: urban populations stable-to-increasing, rural populations stable-to-declining, with regional recovery from the early-2000s West Nile virus impact uneven across geography.

Read more inCrow population trends in North America: what acoustic monitoring showsAmerican crow population trends · crow population decline · West Nile crow recovery · eBird crow data

How did West Nile virus affect crow populations?

Severely. Regional populations declined by thirty to seventy percent across affected areas of the eastern and midwestern United States from 2003 through about 2007. Crows have substantially lower West Nile resistance than many adjacent bird species. Recovery has been partial; some regions have rebuilt to pre-2003 levels, others haven't.

Read more inCrow population trends in North America: what acoustic monitoring showsAmerican crow population trends · crow population decline · West Nile crow recovery · eBird crow data

Are American crows endangered?

No. The species is not currently considered of conservation concern at federal or international levels. Populations are large, distribution is broad, and the species is highly synanthropic — adapting well to human-modified landscapes.

Read more inCrow population trends in North America: what acoustic monitoring showsAmerican crow population trends · crow population decline · West Nile crow recovery · eBird crow data

What is the honeybee waggle dance?

A stereotyped figure-eight dance performed by honeybee foragers returning to the hive. The direction of the dance's 'waggle' segment encodes the direction to a food source relative to the sun; the duration encodes the distance. Other bees observe and fly to the indicated location. Karl von Frisch's research on the dance won the 1973 Nobel Prize.

Read more inHoneybee waggle dance: the gold standard for animal communication decodinghoneybee waggle dance · Karl von Frisch dance language · animal communication decoding · decoded animal language

Is the honeybee waggle dance considered animal language?

Not in the linguistic sense. It encodes spatial coordinates and food-quality information, not lexical or compositional content. The closest fair characterization is 'a quantitative spatial map projection.' Calling it a 'language' overreaches what's encoded — but the decoding is rigorous in ways no animal language has matched since.

Why is waggle dance the 'gold standard' for animal communication decoding?

Three properties together: rigorous sender-side encoding (waggle parameters quantitatively map to spatial parameters), demonstrated receiver-side decoding (observing bees fly to predicted destinations), and confirmed causal encoding-decoding via manipulation experiments. Most claimed animal communication 'decoding' fails at least one of these three tests; waggle dance passes all three with high confidence.

Could Alex the parrot really talk?

Alex used English-language labels in cognitively-meaningful ways: identifying objects, colors, shapes, and small numbers under controlled testing, including for novel stimuli. He did not demonstrate human-like grammar, sentence-level utterance comprehension, or anything resembling full natural-language use. The accurate framing is 'cognitive label use in a non-primate vocal-learning species,' which is remarkable but distinct from 'talking' in the full human sense.

Read more inAlex the parrot and the comparative-cognition traditionAlex the parrot · Irene Pepperberg · African grey parrot intelligence · parrot communication

Who was Irene Pepperberg?

An American comparative-cognition researcher who studied Alex (an African grey parrot) from 1977 until his death in 2007. Her work established the model/rival training method and documented parrot cognitive capacities including categorical labeling, cross-category questioning, and small-integer counting. She is widely considered one of the founders of modern comparative animal cognition.

Read more inAlex the parrot and the comparative-cognition traditionAlex the parrot · Irene Pepperberg · African grey parrot intelligence · parrot communication

What does the Alex research tell us about crows?

Parrots and corvids are both vocal-learning species with comparable cognitive capacity claims. The Alex methodology establishes a rigor baseline for controlled comparative-cognition work that crow research can aspire to. The under-claiming discipline (Pepperberg never claimed Alex had language despite obvious popular pressure) is exactly the discipline that careful corvid research and AI bioacoustic research need to maintain.

Read more inAlex the parrot and the comparative-cognition traditionAlex the parrot · Irene Pepperberg · African grey parrot intelligence · parrot communication

What is Project CETI?

The Cetacean Translation Initiative, launched in 2020 with approximately $33 million in funding from the Audacious Project. Goal: apply modern AI to sperm whale communication, focused on a population around Dominica in the Caribbean. Combines bio-logging, hydrophone arrays, AI analysis, and behavioral observation. The most ambitious animal-communication research program ever launched.

Read more inProject CETI and the sperm whale codasProject CETI · sperm whale codas · cetacean translation initiative · whale communication AI

Has Project CETI translated whale language?

No, and the project is careful not to claim translation. The work has identified structural properties in sperm whale coda sequences (combinatorial patterns, clan-specific variation, machine-learning-recoverable structure) that are substantively new findings. Whether the species's communication includes referential or compositional content remains an empirical question the project is helping address; the answer might be partial or limited.

Read more inProject CETI and the sperm whale codasProject CETI · sperm whale codas · cetacean translation initiative · whale communication AI

Could similar AI research be applied to crows?

In principle yes; in practice the species-specific traits that make sperm whales tractable (bounded populations, long-term individual tracking, discrete communication units) are mostly absent for American crows. A CETI-equivalent for crows would need dedicated multi-year infrastructure at a budget scale far beyond anything currently funded. CrowLingo's reference atlas is small-scale education and reference work, not equivalent to CETI's primary-research effort.

Read more inProject CETI and the sperm whale codasProject CETI · sperm whale codas · cetacean translation initiative · whale communication AI

Who is Nicola Clayton?

Professor of Comparative Cognition at the University of Cambridge. Her research on western scrub-jays starting in the late 1990s demonstrated episodic-like memory in a non-primate species, reshaping what cognitive science thought was possible in non-human animals. She is one of the most-cited contemporary corvid cognition researchers.

Read more inNicola Clayton and the scrub jay memory studiesNicola Clayton scrub jay · episodic memory in birds · what-where-when memory · western scrub jay caching

Do birds have memory?

Yes, and at least one species (western scrub-jay) shows what-where-when memory — the integration of object identity, location, and temporal information to make adaptive decisions. Whether this constitutes the same conscious recall humans report having is uncertain, but the behavioral signature of episodic-like memory is robustly demonstrated. Related capacities including future-oriented planning have been shown in other corvid species.

Read more inNicola Clayton and the scrub jay memory studiesNicola Clayton scrub jay · episodic memory in birds · what-where-when memory · western scrub jay caching

Why is corvid cognition research focused on caching species?

Cache management requires integrating what-where-when memory under ecologically realistic conditions, which makes caching species (scrub-jays, nutcrackers, some other corvids) natural model systems for memory research. American crows cache less than scrub-jays, which is partly why the species's contribution to memory research is smaller — but other corvid cognition findings (face recognition, social learning, vocal communication) are well-represented in the American crow literature.

Read more inNicola Clayton and the scrub jay memory studiesNicola Clayton scrub jay · episodic memory in birds · what-where-when memory · western scrub jay caching

Do birds have a neocortex?

Not in the mammalian six-layer laminar sense. Birds have a pallium — the same evolutionary structure that mammalian neocortex develops from — organized in columnar functional units comparable to mammalian cortical columns. The 2020 Stacho paper established that avian pallium has cortex-like organization at the cellular level. Bird brains are functionally analogous to mammalian brains in ways the older anatomy didn't recognize.

Read more inThe avian pallium and the bird brain revolutionavian pallium · bird brain neocortex · Stacho 2020 · Güntürkün bird cognition

Why did people think bird brains were simple?

Pre-2000s neuroanatomy classified bird forebrain as predominantly basal ganglia based on stained anatomical sections that looked structurally different from mammalian neocortex. The classification was wrong: bird forebrain is predominantly pallium, the evolutionary cortex equivalent. The 2002 Avian Brain Nomenclature Consortium paper reclassified the structures; the 2020 Stacho paper established the functional column organization.

Read more inThe avian pallium and the bird brain revolutionavian pallium · bird brain neocortex · Stacho 2020 · Güntürkün bird cognition

What does the new neuroanatomy mean for crow intelligence claims?

Behavioral findings suggesting corvid cognition approaches primate cognition in some domains used to face an anatomical objection ('but bird brains don't have neocortex'). The updated anatomy removes that objection. The cognitive substrate that could support sophisticated communicative behavior is present in corvid brains. What the communication system actually does with that substrate is the empirical question AI bioacoustic research is investigating.

Read more inThe avian pallium and the bird brain revolutionavian pallium · bird brain neocortex · Stacho 2020 · Güntürkün bird cognition

Will AI translate animal language by 2030?

Almost certainly not in the human-language-translation sense. AI bioacoustic research will probably produce substantial structural mapping of communication systems in specific species (Project CETI is the leading example for sperm whales) and may produce richer understanding of structural properties like combinatorial patterns and individual signatures. None of this is 'translation' in the human-language sense, and serious researchers in the field are careful not to claim it will be.

Read more inThe next decade of bioacoustic AI: a careful forecastfuture of bioacoustic AI · AI animal language 2030 · bioacoustics next decade · Perch NatureLM future

What will AI bioacoustics actually achieve in the next decade?

Practical conservation applications: PAM-network-based population monitoring, acoustic biodiversity assessment, early detection of population declines. Improved foundation models with broader species coverage and cross-species generalization. Real-time species identification at high accuracy from low-cost devices. More-than-superficial structural understanding of communication systems in a few well-funded study species. Substantial scientific progress without crossing into 'animal language translation' territory.

Read more inThe next decade of bioacoustic AI: a careful forecastfuture of bioacoustic AI · AI animal language 2030 · bioacoustics next decade · Perch NatureLM future

What about American crow research specifically?

Plausible developments include standardized PAM networks producing long-term acoustic data on the species, foundation-model embeddings supporting cross-population dialect comparison, wearable-logger studies extending the Demartsev carrion crow methodology, and open-corpus public-reference projects (CrowLingo is one early example). The pace depends on research funding and infrastructure deployment, both of which are smaller-scale than for cetacean research currently. The trajectory is plausible but not certain.

Read more inThe next decade of bioacoustic AI: a careful forecastfuture of bioacoustic AI · AI animal language 2030 · bioacoustics next decade · Perch NatureLM future

Are crows as smart as dolphins?

On many cognitive measures yes, with the caveat that 'smart' is a multi-dimensional concept and direct comparison is difficult. American crows show face recognition, social information transfer, individual signature recognition, future-oriented behavior, and vocal-learning capacity at levels comparable to dolphins on equivalent measures. Both species are part of the convergent-intelligence pattern across distantly-related lineages, with different neural architectures producing similar cognitive capacities.

Read more inConvergent intelligence across distant lineagesconvergent evolution intelligence · octopus crow dolphin elephant · comparative animal cognition · evolution of intelligence

Why do so many distantly-related animals show high intelligence?

Convergent evolution: similar ecological pressures (complex social environments, varied food resources, long-term life-history strategies) favor intelligence in multiple lineages, and intelligence can be supported by multiple neural architectures (mammalian cortex, avian pallium, cephalopod distributed nervous system). The convergence suggests intelligence is selectively favored under recognizable conditions and is achievable via multiple biological pathways.

Read more inConvergent intelligence across distant lineagesconvergent evolution intelligence · octopus crow dolphin elephant · comparative animal cognition · evolution of intelligence

Does convergent-intelligence research undermine the 'human intelligence is unique' view?

It reframes it. The cognitive capacity cluster appears repeatedly in evolution; humans have an unusual elaboration of the capacities (particularly in language, abstract reasoning, cumulative cultural elaboration) but the underlying machinery is not species-unique. The reframing changes what makes humans distinctive: not presence of certain capacities, but degree and combination of capacities elaborated beyond non-human levels.

Read more inConvergent intelligence across distant lineagesconvergent evolution intelligence · octopus crow dolphin elephant · comparative animal cognition · evolution of intelligence

What does Indigenous knowledge contribute to corvid research?

Pacific Northwest Indigenous traditions encode substantial knowledge of raven and crow behavior accumulated over millennia: social structure, intelligence, individual recognition, seasonal patterns, behavioral details. Many findings that Western ornithology has documented recently were already present in Indigenous knowledge frameworks. The knowledge systems aren't equivalent or directly translatable, but they overlap in substantial ways.

Read more inIndigenous knowledge and Western corvid scienceindigenous knowledge corvid · traditional ecological knowledge crow · raven mythology Pacific Northwest · indigenous bird knowledge

What is two-eyed seeing?

An approach to knowledge coined by Mi'kmaq elder Albert Marshall (Etuaptmumk in Mi'kmaq) that draws on Indigenous knowledge through one eye and Western scientific knowledge through the other, using both together for fuller understanding. Increasingly influential in ecology, conservation biology, and beginning to enter corvid research at early stages.

Read more inIndigenous knowledge and Western corvid scienceindigenous knowledge corvid · traditional ecological knowledge crow · raven mythology Pacific Northwest · indigenous bird knowledge

Does CrowLingo represent Indigenous knowledge?

No, the atlas is positioned within Western scientific bioacoustic research and doesn't represent Indigenous knowledge systems. The atlas's content overlaps with topics where Pacific Northwest Indigenous knowledge is substantial and authoritative; future expansion could plausibly include collaboration that respects both traditions. The current scope is one knowledge system; acknowledging this limit explicitly is part of accurate framing.

Read more inIndigenous knowledge and Western corvid scienceindigenous knowledge corvid · traditional ecological knowledge crow · raven mythology Pacific Northwest · indigenous bird knowledge

Do vervet monkeys really have words for predators?

They have alarm calls that function as referential signals — meaning the calls reliably indicate specific predator categories and produce category-appropriate behavioral responses in receivers. The Cheney-Seyfarth playback experiments demonstrated this rigorously. Whether this constitutes 'words' in the human linguistic sense is contested; the careful framing is 'referential signaling,' which is demonstrated, rather than 'words,' which overstates the linguistic structure.

Read more inVervet monkey alarm calls: the other gold-standard studyvervet monkey alarm calls · Cheney Seyfarth research · primate alarm call semantics · referential signaling animals

Why is the vervet research considered foundational?

It's one of the cleanest cases of demonstrated referential signaling in any non-human animal. The methodology — sender-side characterization plus receiver-side playback experiments under controlled conditions — set a standard for what rigorous animal-communication research looks like. The Cheney-Seyfarth work approaches the rigor of von Frisch's honeybee waggle dance research within vertebrate research.

Read more inVervet monkey alarm calls: the other gold-standard studyvervet monkey alarm calls · Cheney Seyfarth research · primate alarm call semantics · referential signaling animals

Have crows been tested with similar methodology?

Partially. American crow alarm calls show some discrimination across predator types in field research, but the rigorous playback-experiment work at the Cheney-Seyfarth bar hasn't been done as extensively for crows. Adapting the methodology to crow research would be one of the higher-value next steps for the field, and the methodology is well-established enough to be directly applicable.

Read more inVervet monkey alarm calls: the other gold-standard studyvervet monkey alarm calls · Cheney Seyfarth research · primate alarm call semantics · referential signaling animals

Section

Journal · Ethics

15 Q

What we should and shouldn't claim about AI-decoded animal communication, captive research ethics, decoding limits — from the long-form journal.

What are playback experiments in bioacoustics?

Field experiments in which researchers play recorded vocalizations back to wild animals to test how they respond. Used to test semantic and functional claims about communication. Foundational to modern bioacoustics, but the source of nearly every preventable harm in the field if poorly designed.

Read more inThe ethics of playback experiments in bioacoustic researchbioacoustic ethics · playback experiments ethics · IACUC bird research · wildlife welfare

Why doesn't CrowLingo do playback experiments?

CrowLingo is a public-facing editorial site, not a research lab. We don't have IACUC approval, don't conduct field experiments, and don't deploy playback in the wild. A million users with backyard Bluetooth speakers playing alarm calls is not equivalent to a million ethics-reviewed experiments; the harms compound.

Read more inThe ethics of playback experiments in bioacoustic researchbioacoustic ethics · playback experiments ethics · IACUC bird research · wildlife welfare

What rules govern responsible bioacoustic playback?

In the US, IACUC review for any vertebrate-wildlife playback. Per-location exposure typically capped under ninety seconds per week, no playback within ten meters of active nests, no alarm or distress calls during breeding season, synchronized observation to catch distress early, and immediate termination if signs of harm appear.

Read more inThe ethics of playback experiments in bioacoustic researchbioacoustic ethics · playback experiments ethics · IACUC bird research · wildlife welfare

Is there a replication crisis in animal cognition?

The field hasn't had the same systematic replication-checking work as psychology, but enough high-profile findings have failed to replicate at the strength of original claims to suggest the issues are present. Animal cognition is structurally hard to replicate (small samples, species-specific setups, expensive infrastructure), so the picture is harder to assess than in lab psychology, but caution about over-strong claims is warranted.

Read more inThe replication problem in animal cognitionreplication crisis animal cognition · animal intelligence research · comparative cognition reproducibility · primate research replication

Have crow cognition findings replicated?

Most of the central corvid-cognition results have replicated robustly — Marzluff's face recognition work, Clayton's episodic-like memory studies, Pepperberg's Alex categorical labeling work all have multiple independent confirmations. Some less-central or more recent findings have less replication evidence and warrant appropriate epistemic caution.

Read more inThe replication problem in animal cognitionreplication crisis animal cognition · animal intelligence research · comparative cognition reproducibility · primate research replication

What does this mean for AI bioacoustic research?

Most recent AI bioacoustic findings haven't been systematically replicated yet — the field is too new. Some will replicate robustly, some will be revealed as artifacts, some as 'smaller than initially claimed.' This is the normal trajectory for early-stage research areas. Careful framing distinguishes what's established from what's preliminary. CrowLingo's atlas adopts confidence-honest framing precisely to avoid contributing to over-claim.

Read more inThe replication problem in animal cognitionreplication crisis animal cognition · animal intelligence research · comparative cognition reproducibility · primate research replication

Can AI translate crow vocalizations into English?

Almost certainly not, regardless of how much data or compute is applied. The verification problem (you can't ask a crow what it meant), the non-existence problem (if the underlying communication isn't compositional, decoding it as compositional finds patterns that aren't really there), and the cognitive-content problem (knowing which call goes with which context doesn't tell you what the crow is 'thinking') are structural limits that don't dissolve with better methodology. Careful researchers in the field don't claim translation will be achieved.

Read more inWhat machine learning probably can't decode, no matter how much audio we feed itlimits of AI animal communication · what AI cannot decode · machine learning animal language limits · untestable animal cognition claims

What can AI bioacoustic research deliver?

Structural mapping of communication systems, statistical relationships between vocalizations and behavioral contexts, identification of individual signatures and dialect patterns, embedding-model analyses that reveal acoustic-similarity geometry, and increasing precision in characterizing what species produce vocally. All scientifically substantive; none of it 'translation' in the human-language sense. The careful framing supports progress that survives critique.

Why is the careful framing so important?

Research framing that promises what methodology can't deliver sets the field up for disappointment, public backlash, and funding cuts when promises don't materialize. The careful framing isn't a weakness; it's a research-program design feature that allows accumulated findings to support real understanding without crossing into over-claim. The honest version of where this field can go is genuinely interesting and substantial.

Does CrowLingo translate crow vocalizations?

No, and the atlas is explicit about not doing this. The behavioral-context probability bars describe statistical associations between vocal patterns and observed behavioral contexts; they don't claim those associations constitute meaning. The 'we don't claim translation' positioning is editorial discipline grounded in the underlying research's actual capabilities.

Read more inWhat the CrowLingo atlas is notCrowLingo limitations · what CrowLingo doesn't do · atlas boundaries · scope limitations bioacoustic atlas

Is CrowLingo a substitute for academic literature?

No. The atlas is a synthesized public-reference work for engaged general audiences, students, science journalists, and conservation practitioners. Specialist research should reference original academic sources directly. CrowLingo's library and citations point to those sources rather than replacing them.

Read more inWhat the CrowLingo atlas is notCrowLingo limitations · what CrowLingo doesn't do · atlas boundaries · scope limitations bioacoustic atlas

What does CrowLingo do?

Provides an accessible editorial atlas of American crow vocal communication, grounded in what current AI bioacoustic research has established, with careful framing about confidence levels and outstanding questions. Synthesizes research from multiple labs and projects. Maintains an open-corpus reference. Supports engaged observation by general audiences. Doesn't claim to do more than this.

Read more inWhat the CrowLingo atlas is notCrowLingo limitations · what CrowLingo doesn't do · atlas boundaries · scope limitations bioacoustic atlas

Is captive corvid research ethical?

Contested, with two main positions. Animal-welfare advocates argue it can be ethically acceptable provided welfare standards are sufficient (housing, enrichment, social grouping, voluntary participation). Animal-rights advocates argue captivity for research is fundamentally problematic regardless of welfare. Most academic ethics frameworks operate on welfare principles. Reasonable people disagree.

Read more inCaptive corvid research: what the ethical debate involvescaptive corvid research · animal research ethics · corvid welfare research · captive bird ethics

Why are corvid welfare considerations specifically challenging?

Cognitive complexity makes solo housing especially problematic. Behavioral flexibility makes environmental enrichment important. Long lifespan (15-30+ years) raises lifetime-of-research considerations. Capacity for what looks like distress at conspecific death raises questions about handling losses in captive populations. Modern well-resourced corvid research has developed protocols for these; older work often didn't.

Read more inCaptive corvid research: what the ethical debate involvescaptive corvid research · animal research ethics · corvid welfare research · captive bird ethics

Is the field moving away from captive research?

Partly. Trends include more wild-population observational work (Demartsev wearable loggers), more cooperative captive research where birds participate voluntarily, longer-term lifetime considerations in captive programs, and welfare integrated with cognition research. Some research (complex experimental manipulation, controlled-condition cognition tests) will probably continue to require captive populations.

Read more inCaptive corvid research: what the ethical debate involvescaptive corvid research · animal research ethics · corvid welfare research · captive bird ethics

Section

Home

3 Q

Reference and editorial questions about home.

What is CrowLingo?

CrowLingo is a public-facing exploration of what AI audio foundation models reveal about American crow vocalizations. It pairs an interactive vocal atlas, real CC-licensed audio recordings, real spectrograms, and AI-narrated cluster explanations grounded in primary corvid literature.

Is CrowLingo translating crow language?

No. The site catalogs and characterizes crow vocalizations using audio embeddings and behavioral context — it does not claim to translate. The ethics floor explicitly rules out translation claims; interpretation is acknowledged as interpretation.

Who built CrowLingo?

CrowLingo is a Kymata Labs publication. Audio sources are CC-licensed from Wikimedia Commons (primarily Jonathon Jongsma, CC BY-SA 3.0). AI narration is grounded in the published literature listed in the library.

Library

1 Q

Reference and editorial questions about library.

What references back CrowLingo's claims?

Twelve primary references organized into four constellations: foundational corvid behavior (Marzluff, Heinrich, Mates, Verbeek, Demartsev), bioacoustic methods (Stowell, Bradbury & Vehrencamp, Kahl), SSL & foundation models (Robinson/NatureLM, Hagiwara/BEANS), and animal-language AI ethics (Earth Species Project, Andreas/CETI). Each entry links to its canonical source and has a 30-second AI narration.

Reference

4 Q

Reference and editorial questions about reference.

What is an audio embedding in bioacoustic AI?

An audio embedding is a learned vector representation of an input audio clip — for crow audio specifically, BirdNET produces a 1,024-number list per clip and Perch 2.0 produces a 1,536-number list. The embedding captures acoustic content (frequency profile, spectral structure, temporal pattern) richer than any hand-engineered feature, learned automatically from massive unlabeled training data via self-supervised objectives. The geometry is what makes the representation useful: similar-sounding clips produce nearby vectors, dissimilar clips produce distant vectors, and downstream tasks like species classification, individual identification, or behavioral-context clustering operate on those distances. Modern bioacoustic AI is fundamentally an embedding-based field; BirdNET, Perch, and NatureLM-audio are all embedding models at their core.

Read more inGlossary

What is UMAP (Uniform Manifold Approximation and Projection) and why is it used for crow vocal atlases?

UMAP is a non-linear dimensionality-reduction algorithm that flattens high-dimensional embeddings (1,024-dim or 1,536-dim audio vectors) down to two or three dimensions for human-inspectable visualization, while preserving local neighborhood structure — meaning points that are similar in the original high-dim space stay close in the 2-D projection. UMAP replaced earlier methods like t-SNE in most bioacoustic visualization work because it preserves both local and global structure better and runs at larger data sizes. CrowLingo's vocal atlas is a UMAP projection of NatureLM-audio embeddings; the cluster geometry visible in the 2-D map approximates real similarity geometry in the underlying high-dim representation. Similarity search itself always runs on the full-dim embedding, never on the 2-D UMAP coordinates.

Read more inGlossary

What is HDBSCAN clustering and how does it find vocal categories without specifying a target count?

HDBSCAN (Hierarchical Density-Based Spatial Clustering of Applications with Noise) is a density-based clustering algorithm that finds dense regions in high-dimensional space without requiring the user to specify the target cluster count in advance. The algorithm builds a hierarchical structure of density-connected points, identifies stable clusters at multiple density thresholds, and outputs cluster assignments plus a noise label for points that don't fit any cluster confidently. This is exactly the property you want for animal vocal repertoires — you don't know in advance how many vocal categories a species has, you want the data to tell you. CrowLingo's atlas uses HDBSCAN on the full-dim embedding to discover the nine clusters of American crow vocalizations; the cluster boundaries are data-driven, the cluster labels (territorial, mobbing, etc.) are subsequently assigned by matching against the prior descriptive literature.

Read more inGlossary

What is self-supervised learning in audio and why does it matter for bioacoustics?

Self-supervised learning is a training paradigm where a model creates its own supervision signal from unlabeled data — for audio, typically by predicting masked or hidden portions of a spectrogram from the surrounding context, or by learning that two clips from the same source should have similar representations while clips from different sources differ. The model never sees an explicit human label during pretraining; the supervision emerges from the structure of the data itself. This matters for bioacoustics because labeled wildlife audio is scarce and expensive, while unlabeled wildlife audio exists in enormous quantities (Macaulay Library has 1.3M+ recordings). Self-supervised pretraining on the large unlabeled corpus produces rich representations that downstream tasks can use with much less labeled fine-tuning data. BirdNET, Perch 2.0, and NatureLM-audio are all built on this paradigm; it's how the modern bioacoustic AI revolution actually works.

Read more inGlossary

Didn’t find your question? Try the journal or the atlas.

80 long-form essays