Four ideas that change how we listen to animals.
A dog breathing fast after a walk sounds harsher than the same dog at rest. This is normal. The instrument knows the difference. It compares harshness against what is expected for this exact breathing rate, breed, age, and weight — not a generic average.
Every phone microphone is different. An iPhone filters sound differently from a Samsung. The instrument does not fight this. It uses the same phone to compare today's recording against last week's. The phone's own distortion cancels out mathematically. No calibration required.
First, the instrument recognises patterns in the sound using thousands of reference recordings from veterinary databases. Then, a deterministic validator checks whether the recognition makes physical sense. When both agree: high confidence. When they disagree: the instrument admits uncertainty.
Most systems guess. This instrument refuses. If the signal is too noisy, too brief, or too ambiguous, it says so. It withholds metrics rather than invent them. It flags "unclassifiable" rather than force a wrong label. Honesty is a feature, not a bug.
Fourteen synthetic tests with known inputs assert measured precision. External validation against seventeen reference datasets. The instrument spec sheet is published, not claimed.