Neigh.
It tells you what it heard. In a voice you choose, and it tells you when it could not tell.
What listening does
Put the phone down near the stable and walk away. It counts contact calls while you are gone, which is the one thing you cannot hear for yourself. Nothing is recorded unless you save a clip yourself: it listens, counts, and forgets the audio.
Catch mode is for the sound that just happened: it holds only the last 30 seconds in memory. “Catch that” reads the most recent 15 of them; “Save last” keeps 15 or 30 as a file you can post. Walk away without tapping either and nothing was kept.
Talk to a horse
Horses have a handful of calls, each with a job. Pick one and it plays the real recording. This is a phrasebook, not a translator: it speaks the horse's actual calls, not your sentences turned into horse.
Or type what you want to say
Most sentences have no horse equivalent, and it will say so. That is honesty, not a failure.
How it listens, and how well
Every number below was measured on independent recordings that people, not this app, labelled. Most apps in this category publish no numbers at all, and there is a reason.
Measured August 2026: horse sounds at 94% precision (95% CI 85–98%) and 71% recall (CI 61–79%) on 148 randomly drawn clips people labelled by ear, held out of every tuning decision; whinny naming at 65% precision (CI 49–78%) and 92% recall (CI 76–98%) on the same draw. The ranges are shown because a number without its range is a guess in a suit. Missing is normal: leave it listening and repeated calls make the odds add up.
Everything runs on this phone. A clip is uploaded only when you press Share on it and confirm, and the confirmation screen lists exactly what leaves.
How we measure, so you can check us — every claim, the test behind it, and the rules the numbers live under.
The dictionary, and how it grows
Every horse call earns its phrase the same way: measured against sounds people labelled, and silent until it clears the bar. This is where each one stands.
The nicker has no honest score to publish yet. The 8-in-10 figure this card used to show came from labels taken off video titles, and when people actually listened to those clips most of the nickers were not nickers — several were human voices. Both numbers have to clear the bar on sounds people have confirmed by ear, and until enough of those exist there is nothing here worth quoting.
The nicker recognizer is learning from public recordings and shared sounds. It stays silent in the app until it clears the same bar the whinny cleared, because a wrong word in your horse's mouth is worse than none. Sharing a sound is what moves it, and a sound you have confirmed by ear moves it most.