Stuck? Write to us. A person reads it.

Because it genuinely didn't, and we'd rather tell you than invent something.

Before any reading happens, PawParley strips the silence out of your recording and checks two things: whether enough sound survived to be worth measuring, and whether what survived resembles a cat or a dog at all. If either check fails, you get nothing — no guess, no filler.

The usual causes are distance, background noise, and timing. Try holding the phone closer, turning off the television, and starting the recording while your pet is making the sound rather than just after. A single very short chirp can also fall below the threshold.

This is the feature we're proudest of, so it isn't going away. It's the reason the app can't behave like a Magic 8 Ball.

Because we don't believe anyone in this category who does, and we're not going to join them.

Accuracy figures for pet vocalisation depend enormously on how you split your test data. We built a mood classifier and measured it honestly, holding each individual cat out of the test set — and it scored below the majority-class baseline. The same model looked respectable when the same cats appeared in both training and testing. That gap is the model memorising individual voices rather than learning meaning, and it's exactly how impressive-sounding numbers get produced.

For context: published work puts human listeners at around 65% on cat call context, and the best six-way bark classification in the literature at 43%. Anyone claiming a lot more than that owes you an explanation of their test split.

So there is no percentage in PawParley. What you get instead is the measurement, and a confidence indicator that drops when the evidence is thin.

No — and nor can anyone else, including us. Anything sold as a cat translator is guessing.

The reason is structural rather than technical. Researchers who study cat vocalisation describe the meow as a signal learned between one cat and one person, not a vocabulary shared across the species. Your cat's feed me is not the cat next door's feed me. There is no dictionary, so there is nothing to look a word up in.

What can be done honestly is measurement. A meow lasts about 0.42 seconds on average in the published descriptions, its fundamental frequency typically falls somewhere between 221 and 1185 Hz, and its pitch travels in a direction. That direction is the most useful cue anyone has found: across the cat literature, calls made when a cat wants something are wide and rising — a pitch range around 251 Hz — while calls made under stress are narrow and flat or falling, around 119 Hz. Human listeners asked to judge which situation a meow came from score roughly 65%.

So PawParley measures duration, pitch contour, how far the pitch moved and how the sound repeated, then tells you what that resembles. It never claims to know what your cat said.

Adult cats barely meow at each other at all. The meow is aimed at humans.

Among themselves cats communicate mostly through scent, posture and tail position, plus a small close-range set of sounds — the hiss, the growl — which are about conflict rather than conversation. The meow sits outside that: it is a channel cats developed to reach people, which is why it is the sound you hear most and the one your cat aims squarely at your face.

The other sound worth knowing is the trill, or chirrup — half a second or so, made with the mouth closed, rising, somewhere around 358 to 533 Hz. It is the sound researchers read most consistently as a friendly greeting, and it is the most reliably positive thing in the whole cat repertoire.

Yes. The reading is derived from the audio itself, so an identical recording produces an identical result every time. There is no random draw anywhere in it.

Two different recordings of what sounds to you like the same meow can land differently, because they aren't acoustically identical — different distance, different room, a slightly different call. That's the measurement being honest, not the app being random.

Less than the marketing on other apps would suggest, and we'd rather be straight about it.

Dogs hear from roughly 67 Hz to about 45 kHz. A real ultrasonic dog whistle works up at 23–54 kHz. A phone physically cannot produce that. The audio pipeline tops out at 24 kHz in theory, and the speaker itself rolls off sharply above about 19–20 kHz and collapses past 21. No app can code its way around the hardware.

What a phone can do is produce a clean tone in the 11–20 kHz region, which is comfortably inside canine hearing and close to inaudible for most adults. That's what PawParley's whistle does, defaulting to a 12–16 kHz training tone, with the top of the dial marked as reduced output rather than quietly pretending.

It is a training aid — a distinct, consistent sound you can pair with a behaviour. It is not a device that stops barking, and we won't sell it as one. Use it at a sensible volume, never right by your pet's head, and keep sessions short.

Nowhere. All the analysis happens on your iPhone, the models are bundled inside the app, and no audio is ever uploaded — not to us, not to anyone. Anything you save to the diary is stored in the app's own storage on your device and is deleted with the app.

There's no account to make and nothing to sign in to. The full detail is in the Privacy Policy.

Yes. The sound reader, the whistle, the clicker and the diary all work with no network connection at all, because none of them need one. The only thing that needs the internet is buying or restoring a subscription, which goes through the App Store.

Because the research is. Cat calls have one reasonably strong published cue — how far the pitch moves and which way it goes. Dog barks have real, measurable structure too, especially in how fast they repeat, but the literature disagrees with itself about what a positive bark sounds like, and barks from dogs left alone measure almost identically to barks from dogs at play.

So the dog side hedges harder and shows lower confidence more often. That's the honest position, not a missing feature.

You can point it at anything, but it will almost certainly tell you it didn't hear a cat or a dog — which is the correct answer. The sound identification is built for two species and doesn't pretend otherwise.

No, and please don't use it that way. PawParley is entertainment. It does not diagnose, it will never tell you an animal is in pain or unwell, and no reading it gives you should influence a decision about your pet's health.

If something seems wrong with your animal, ring your vet.

Every sound in the app has a visual equivalent, so nothing is lost with the volume down. If something is hard to use with VoiceOver, at a larger text size, or with reduced motion switched on, tell us — we treat those as bugs, not requests.

hello@pawparley.com