What changed under the hood, and why crows are the right test species.

For fifty years, bioacoustics ran on hand-labeled spectrograms, a handful of named call types per species, and the slow grind of a graduate student's ear. Self-supervised audio models, latent-space analysis, and foundation models like NatureLM-audio changed the floor of the field around 2022–2025.

The leverage for crows specifically: a single species with a dense, graded repertoire, well-documented individual recognition, and decades of behavioral context to compare against. Methods that look like party tricks elsewhere become load-bearing here.