A game with real sounds
What a cat's meow says: 440 real meows put to the test
Researchers at the University of Milan clipped a small microphone to the collars of 21 cats and recorded them in three situations: waiting for food, being brushed, and left alone in a strange room. People listening to unfamiliar cats get about one in three right. Can you?
The sound comes from the collar microphone, so it sounds like a phone call. Play it as often as you like. The picture is the meow itself: time runs left to right, higher notes sit higher.
Your score
The thin line is blind guessing: one in three. The people in the study heard six meows recorded by the same team, three picked as typical and three as unusual; the bar shows only the typical ones, so it is a reference point, and the sounds need not be the ones you heard. The model's score is not clearly distinguishable from blind guessing.
What people hear
How well do people understand an unfamiliar cat?
In 2020 the same Milan team asked 225 adults to listen to six meows it had recorded from the ten Maine Coons of a single owner: one typical and one unusual meow for each situation. The typical ones were identified correctly 40% of the time for food, 33% for brushing and 27% for the strange room. Taken together, 33%: exactly what blind guessing would give with three options.
Two of the unusual ones came out well below chance: 13% for brushing and 14% for the strange room. There the sound itself pulled people towards a wrong situation. The food one, at 27%, was about as low as the typical strange-room meow (27%). Women and cat owners did better on brushing and the strange room.
Other studies agree that people do poorly. In 2003, 28 listeners sorted 100 meows from 12 cats into five situations and got 27% right, against 20% by chance: better than chance, but not by much. And in 2015, ten owners listened to meows from their own cat and from a stranger's: four of them beat chance on their own cat, none on the unfamiliar one.
What about a computer?
The same 440 meows also went through a simple model that measures the timbre, pitch and length of each sound. Tested on a cat it had never heard, it named the situation 40% of the time, averaged over the three situations, against 33% by chance. Across 200 shuffles of the labels, with each burst of meows kept together, a margin that large turned up by luck in 11 of them. It is no proof that it understood anything.
The people who made the recordings reported up to 96% accuracy in 2019, using ten-fold cross-validation, without saying whether the test cats were kept out of training. With this model, a random split of the meows gives 52%; holding the test cat out of training drops it to 40%. That suggests part of what a computer "learns" may be the recording session and the cat's own voice and home.
In these recordings, a meow shows who is meowing far more clearly than the situation the cat is in.
A cat you know
If you knew the cat, would you guess better?
Owners say they understand their cat. Try a very short version: pick one of three cats, hear two of its meows from each situation, then guess other meows from the same cat, recorded in other sessions a month apart.
Listen to its meows. Each button plays a meow from the situation named beside it.
The model also does a little better once it has heard the same cat in other sessions: 44% instead of 40%. But two meows are no substitute for years spent with a cat.
Map of 440
A map of meows: the ones that sound alike sit together
Each stroke is one meow. Its shape is the tune: it rises where the sound rises, falls where it falls, and the longer the meow, the longer the stroke. Meows that sound alike in timbre, pitch and length sit close together. Tap one to hear it.
Coloured by cat, islands form, and the labelled ones are cats recorded in several situations: relative to chance, a meow's neighbours are more often the same cat in another situation than another cat in the same situation. Some islands, though, are cats recorded almost only in one situation, so the map alone cannot separate the two. The model confirms it: trained on meows from two situations, then asked to identify the cat from meows in the third, which it had never heard from that cat (the 18 cats recorded in all three situations), it named the right cat 31% of the time against 6% by chance, 5.6 times more often. None of the 200 shuffles reached it. Between cats from different homes, though, the room or the fit of the collar can also give the cat away. The test inside a single home removes the differences between homes, though not how each collar sits: the 10 Maine Coons belong to one owner and live in the same conditions. Among the 9 recorded in all three situations, the model got 33%, against 11% by chance.
Why cats meow at us
Why do cats meow at people?
Researchers have not shown the cause directly, but the evidence suggests the meow is a call that draws people's attention and asks for a response. Adult cats rarely meow at each other, a few papers report, and some wild felids meow as kittens and rarely as adults. Hence the hypothesis that the house cat keeps a kitten's call into adulthood and uses it on people; it has not been tested directly. The observations behind it come largely from doctoral theses and a book chapter, and even the vocal African wildcat does not seem to aim its meows at people.
Domestication seems to have changed the call too. A 2004 study compared meows from domestic cats with those of African wildcats, close relatives kept at a zoo: the domestic ones were shorter (0.84 against 1.50 seconds), higher (mean fundamental frequency, 609 against 255 Hz) and sounded far more pleasant to listeners (4.58 against 2.43 on a 1 to 7 scale). The author reads this as selection for what appeals to people, but could not rule out that pets and zoo animals were recorded in different conditions, or that pleasantness was a side effect of selection for other traits.
Learning may contribute as well. In conditioning experiments cats learned to change the pitch, duration and rate of their meows, so a call that brings food or an open door could be repeated. The same author thought it possible that cats learn which calls work on people, and that the house cat is more able or motivated to do so than the wildcat, which despite plenty of contact with people did not seem to; the study did not test whether cats learn to adapt their meows when addressing people. Cats are known to pick out their owner's voice among strangers' voices, and a study published in 2023 found they told speech meant for cats from speech meant for adults only when their owner was talking. The idea that each cat shapes its meows for its own person was put forward in 2003 but has not been tested directly: the closest related test, in 2015, had ten owners.
Your cat
Where would your cat sit on the map?
Press the button when your cat is about to meow and hold the phone close. It records five seconds and keeps the loudest part. The sound stays on your device and is not sent anywhere.
You will see its meow, its place on the map and which of the 21 cats it sounds most like. The page does not tell you what your cat wants: at 40% on unfamiliar cats, that would be a guess dressed up as an answer.
The microphone starts only when you press.
Method and sources
How it was made
The recordings (CatMeows, University of Milan, CC BY-NC 4.0 licence): 440 meows from 21 cats (10 Maine Coons, 11 European Shorthairs), 92 waiting for food, 127 being brushed and 221 in the strange room, captured by a Bluetooth microphone on the collar at 8,000 samples a second. Here they are compressed to AAC, with everything below 120 Hz filtered out; the pictures are computed from the sound you hear, with the steady background noise of each frequency removed from the picture, not from the sound.
The model: a logistic regression on 32 measures of each sound (13 timbre coefficients with their mean and spread, the spectrum's centre of gravity, duration, median pitch, range and direction of the tune, and the share of the sound that is voiced). Scores are averaged over situations (or cats), so the most common class cannot win by size. The chance tests shuffle labels over whole bursts, not single meows. The map is a t-SNE projection of the same 32 measures.
Limits: two breeds, 21 cats, three staged situations, and the strange room also differs from home in its acoustics. A richer model might get more out; the result says what these simple measures can hear.
Sources
- Ludovico, L.A., Ntalampiras, S., Presti, G. et al. (2020). CatMeows: A Publicly-Available Dataset of Cat Vocalizations. Zenodo, CC BY-NC 4.0. zenodo.org/records/4008297
- Ntalampiras, S., Ludovico, L.A., Presti, G. et al. (2019). Automatic Classification of Cat Vocalizations Emitted in Different Contexts. Animals 9(8):543. pmc.ncbi.nlm.nih.gov/articles/PMC6719916/
- Prato-Previde, E., Cannas, S., Palestrini, C. et al. (2020). What's in a Meow? A Study on Human Classification and Interpretation of Domestic Cat Vocalizations. Animals 10(12):2390. pmc.ncbi.nlm.nih.gov/articles/PMC7765146/
- Nicastro, N., Owren, M.J. (2003). Classification of domestic cat (Felis catus) vocalizations by naive and experienced human listeners. Journal of Comparative Psychology 117(1):44–52. doi.org/10.1037/0735-7036.117.1.44
- Ellis, S.L.H., Swindell, V., Burman, O.H.P. (2015). Human classification of context-related vocalizations emitted by familiar and unfamiliar domestic cats: an exploratory study. Anthrozoös 28(4):625–634. doi.org/10.1080/08927936.2015.1070005
- Nicastro, N. (2004). Perceptual and acoustic evidence for species-level differences in meow vocalizations by domestic cats (Felis catus) and African wild cats (Felis silvestris lybica). Journal of Comparative Psychology 118(3):287–296. doi.org/10.1037/0735-7036.118.3.287
- Tavernier, C., Ahmed, S., Houpt, K.A., Yeon, S.C. (2020). Feline vocal communication. Journal of Veterinary Science 21(1):e18. doi.org/10.4142/jvs.2020.21.e18
- Saito, A., Shinozuka, K. (2013). Vocal recognition of owners by domestic cats (Felis catus). Animal Cognition 16:685–690. doi.org/10.1007/s10071-013-0620-4
- de Mouzon, C., Gonthier, M., Leboucher, G. (2023). Discrimination of cat-directed speech from human-directed speech in a population of indoor companion cats (Felis catus). Animal Cognition 26:611–619. doi.org/10.1007/s10071-022-01674-w
CatMeows: CC BY-NC 4.0 · sounds compressed and filtered