What the views mean

Why this is real analysis

The app first asks a simple question: where are the actual sounds? It marks the start and end of each event, measures how strong it is, and checks what frequency range it lives in.

Then it asks whether sounds repeat. If a tone, chirp, call, pulse, or hum keeps showing up, it becomes a contact track. That makes repeated sources visible instead of letting them disappear into the picture.

The app also watches the background. A quiet call inside a quiet recording is different from the same call buried in wind, handling noise, traffic, or water movement.

Underwater audio can use the same ideas when the input is a hydrophone, line-in source, or uploaded recording. A browser cannot magically know direction or distance from one mic, so we call these source tracks, not bearings.

Analysis modules

Class hints are intentionally cautious: tonal, pulsed, mechanical, wind, or low hum. They are clues for review, not final labels.

Compare recordings and annotation mode are future workflows. Comparison belongs in the library, and annotation should let users label contacts or events once the evidence is stable enough to trust.

References

• Robust sound event detection in bioacoustic sensor networks
• Improving Sound Event Detection Metrics: Insights from DCASE 2020
• Clustering and novel class recognition: evaluating bioacoustic deep learning feature extractors
• Global birdsong embeddings enable superior transfer learning for bioacoustic classification
• The SPACE Labelling Tool and Time-Frequency Catalogue workflow
• Pijanowski et al., Soundscape Ecology: The Science of Sound in the Landscape, BioScience, 2011.
• Sharma et al., Contextual and combinatorial structure in sperm whale vocalisations, Nature Communications, 2024.
• Sainburg, Thielk, Gentner, Finding, visualizing, and quantifying latent structure across diverse animal vocal repertoires, PLOS Computational Biology, 2020.
• Hagiwara, AVES: Animal Vocalization Encoder based on Self-Supervision, arXiv:2210.14493.
• Robinson et al., NatureLM-audio: an Audio-Language Foundation Model for Bioacoustics, arXiv:2411.07186.

Library viewer modes

The library page uses a dedicated science-first layout: Playback Path (single active event + transition node), Repertoire Structure (neighbor topology and cluster prominence), and Uncertainty Lens (confidence-weighted node emphasis).