Skip to content
CrowLingo

FIG 7.1 — Glossary · Reference

Glossary.

Definitions of the technical vocabulary used across CrowLingo — bioacoustics, AI audio models, and corvid behavior. Follow a term's deep link to read the guide where it's explained at length.

Embedding
A learned numerical representation of an input. Its size depends on the model, version and selected output; similarity comparisons also depend on the distance measure. An embedding does not translate a call.

Deep dive · Latent space 101 →

Latent space
The vector space containing model representations. Nearby points do not by themselves establish a shared caller, behavior or meaning; the model and distance measure matter.

Deep dive · Latent space 101 →

UMAP
Uniform Manifold Approximation and Projection. A nonlinear dimensionality-reduction method often used to display high-dimensional data in two dimensions; its parameters affect the map.

Deep dive · Latent space 101 →

HDBSCAN
Hierarchical Density-Based Spatial Clustering of Applications with Noise. Finds dense clusters of arbitrary shape without requiring a target cluster count.

Deep dive · Cluster & label →

Self-supervised learning (SSL)
Training paradigm where a model creates its own supervision signal from unlabeled data — e.g. by predicting masked parts of a spectrogram from the rest.

Deep dive · Self-supervised audio →

NatureLM-audio
Earth Species Project's audio-language model for bioacoustics. It learns from paired audio-text data and answers prompts about audio; performance depends on the configuration and evaluated task.

Deep dive · NatureLM-audio →

BirdNET
A family of acoustic models for identifying species and other supported sound classes. Coverage, input settings and embedding size depend on the release.
Perch
Google Research's bioacoustics project, including species classifiers and tools for working with audio embeddings. Check the selected model and task before using its output as evidence.
Syrinx
The bird vocal organ. Located at the bifurcation of the trachea into the bronchi. Two independent sound sources, which is why birds can make two notes at once.

Deep dive · Vocal anatomy →

Spectrogram
A time-frequency-intensity plot of an audio clip. Time on the x-axis, frequency on the y, color = energy at (t, f).
Graded calls
Calls that vary continuously along acoustic dimensions rather than belonging to discrete categories. Appear as bridges between clusters on the vocal map.
Dialect
Consistent vocal differences among groups or regions. Individual differences alone do not establish a dialect. The linked Mates study tested individual information in selected crow calls, not regional or family dialects.

Deep dive · Individuality & dialect →

Syntax (in animal communication)
Structure or rules for combining signals. In compositional syntax, a combination’s meaning derives from its meaningful parts and how they are combined. Predictable call order alone does not establish this; the linked playback example studies great tits, not crows.

Deep dive · Combinatorial evidence →

Sources for model and projection terms

Model and UMAP notes checked against the developers' documentation on November 11, 2026. Select an exact version and output before comparing results; these definitions do not report a shared performance benchmark.

Frequently asked

What people ask about this.

What is an audio embedding in bioacoustic AI?

An audio embedding is a learned numerical representation of a recording or segment. Its size depends on the model and version: BirdNET documents 1,024 dimensions for V2.4 and 1,280 for its V3.0 preview. These are versioned examples, not a universal size for crow recordings. Species scores, embeddings and language answers are different outputs; none establishes a bird's intended message by itself.

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

UMAP is a nonlinear dimensionality-reduction method that can make high-dimensional data easier to inspect in a lower-dimensional plot. The distance metric and parameters such as n_neighbors and min_dist affect the result and the balance of local and broader structure. A plotted neighborhood is not proof of a vocal category or shared meaning. CrowLingo's current Atlas uses illustrative coordinates; no UMAP projection or embedding similarity search has been performed on its corpus.

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 current atlas has nine editorial groups and seeded coordinates, not HDBSCAN results. It illustrates the idea of a repertoire map without claiming nine experimentally discovered categories.

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

Self-supervised learning uses training targets derived from the data itself, such as predicting hidden parts of an input. It can help learn representations for later tasks, but an embedding output does not establish how a model was trained. Training objectives depend on the model and stage: NatureLM-audio's developers describe learning from paired audio-text data. Check the specific model documentation before describing its training or making performance comparisons.