Open tools. Traceable research. Room for wonder. Our apps document observations; animal translation is a research ambition.
AI-created illustration · not a field observation
Birds✳Animals✳Marine life✳Plants✳Fungi
WONDER, WITH A WAY FORWARD
Every signal has a world around it.
Sound becomes more useful when we know who was there, what moved and what happened next. We’re building the tools and the evidence map to connect those pieces.
Animals exchange signals. Plants respond to their environments. Researchers are uncovering patterns we once missed. Our ambition is to help turn curiosity into shared, testable understanding.
We begin with a simple discipline: document what is known, be clear about what isn’t, and build on the work of others. A translation claim needs more than a convincing prediction.
Japanese great tits showed different receiver behaviors under component calls, natural versus reversed order, and single- versus two-source playback; a bounded novel cross-species sequence also elicited stronger responses than its reversal, while a matched ABC–zi control was weak. In the 2022 deposited binary table, the natural single-source 1A-R condition had an approach event in 13/16 flock trials and wing-flicking in 15/16; the other three treatments had approach events in 2–3/16 and wing-flicking in 1–4/16. These aggregate counts reproduce the descriptive pattern only. They do not establish a unique mechanism: one published critique agrees that receiver behavior is sensitive to order and source but disputes that this demonstrates Core-Merge or hierarchical syntax.
NatureLM-audio combines a BEATs audio encoder with Llama 3.1 8B Instruct and responds to audio-plus-text prompts with outputs such as species labels, detections, call types, life-stage classes and captions. Its ICLR 2025 paper reports strong results on several BEANS-Zero tasks and evaluates held-out taxa; the paper cautions that BirdNET and Perch comparison datasets overlap their training data, so those are topline rather than fully fair comparisons. A NeurIPS 2025 workshop follow-up preprint exposes prompt-format brittleness: asking for common and scientific names together reduced accuracy to 6–12% on two benchmark components compared with prompts asking for one name. For those combined-name prompts, the authors report that LoRA interpolation at about α=0.7 raises accuracy from 6% to 45% on Watkins and 12% to 63% on CBI. Separately, on one 425-sample, 40-class closed-set unseen-family task, reported F1 rose from 0.09 to 0.28 (the authors describe this as a >200% relative improvement). These are bounded classification and instruction-following results, not evidence that the model discovers what an animal means or can hold a conversation with it. The current training-data viewer lists 26.4 million rows; the paper notes that prompt counts can derive from the same audio, so row counts are not unique recordings. The official BEANS-Zero card reports 91,965 examples in a single test split and says the composite is not intended for commercial use; it documents a license field per sample. The live Hugging Face tree lists 50.5 GB, while the companion GitHub README says about 100 GB for a complete non-streaming download. Its tasks cover classification, detection, call-type/lifestage and captioning, not receiver comprehension or communication meaning.
Targeted primary-paper and official model/data/code metadata review; no independent reproduction
Audio-language model and benchmark evidenceRead note ↗
Talk2Nature