Cluster · Territorial caw
Perched flock — territorial caws
Perched flock — territorial caws00:25 · 200 Hz – 8 kHz
Perched flock — territorial caws
CrowLingo · Animal language processing · v0
We can finally see what we couldn't hear.
CrowLingo is a focused exploration of one species and one revolution: the American crow, and the new generation of AI audio models that turn its voice into a map.
Listen now
Three calls, three coordinates on the same map.
Each recording is one point in a high-dimensional acoustic space. The full atlas of ~12,400 calls lives at the Repertoire Atlas.
Cluster · Rattle complex
Rattle complex — affiliative
Rattle complex — affiliative00:33 · 200 Hz – 8 kHz
Rattle complex — affiliative
Cluster · Begging
Juvenile begging + adult exchange
Juvenile begging + adult exchange00:40 · 200 Hz – 8 kHz
Juvenile begging + adult exchange
What's actually new
Discrete categories gave way to a continuous map.
placeholder
IG · 02 · BEFORE · AFTER
The fifty-year hand-labeling regime treated crow calls as a handful of named types. Self-supervised audio models flip the frame: every clip becomes a vector, every vector a point in a shared space, every cluster an emergent category. The old labels survive — as labels of regions in the map, not as boundaries on the world.The change is not that we found new sounds. The change is that we stopped treating each call as a label, and started treating the whole repertoire as a geometry. Graded variation, dialect, individual signature — all of it visible at once, on the same map, in milliseconds per call.
The 2026 carrion-crow bioRxiv preprint (Demartsev et al.) used wearable loggers + this mapping discipline to recover both discrete and graded structure in grunts and caws — territory by territory, individual by individual.
Where we are honestly
ALP turned crow vocalization into a map. We have not learned the language.
placeholder
IG · 06 · NOW · VS · AIMED
Demonstrated, today: automatic detection, unsupervised category discovery, caller-identity inference, behavioral-context mapping, zero-shot captioning via NatureLM-audio. Aimed-for, not yet here: compositional decoding, real-time bidirectional dialogue, a 'crow dictionary' with human glosses. The frontier is where the next five years live.Get involved
Four entry points. Pick the one that fits your hour.
Record
Capture a crow well.
Field-rig basics, sampling rate, behavior-sync logging, ≥10 m from active nests.
Learn
What's under the hood.
Self-supervised audio, latent spaces, and the NatureLM-audio foundation model, demystified.
Contribute
Send your recordings.
The open dataset and citizen-scientist pipeline — how to share what you record.
Read
The reading list.
Papers, books, and primary sources, organized by constellation rather than alphabet.