For decades we have fantasized about being like Doctor Dolittle, that character capable of have conversations with any animal Just like that. What was once pure children's literature and Hollywood movies is now beginning to approach reality thanks to artificial intelligence (AI). We're not yet at the point of asking our dog what it thought of the walk or asking a whale to tell us its life story, but science is moving very quickly in that direction.
In recent years, research projects spread across the globe have begun to use advanced algorithms to analyze sounds, gestures and expressions of multiple speciesFrom chickens, pigs, and dogs to whales, dolphins, elephants, and crows, these projects seek something much deeper than a simple "pet translator": they aim to truly understand how animals communicate, improve their well-being, and, in turn, change our relationship with them and the ecosystems we share.
From the fantasy of Doctor Dolittle to AI laboratories
The myth of being able to talk to wildlife is captivating because it connects with a very human desire: to understand what animals feel and think that surround us. Literature already explored this with Doctor Dolittle, a physician who stopped treating people to dedicate himself to animals when he discovered he could speak their language, becoming a naturalist and explorer. It seemed like magic. Today, software is beginning to take on the role of "translator."
The big difference from the past is that now we have deep learning models capable of processing millions of data points audio, image, and context in a very short time. What were once the intuitions of ethologists armed with field notebooks and tape recorders are now becoming enormous digital databases and neural networks that search for patterns invisible to the human ear and eye.
Artificial intelligence doesn't "understand" animals the way a bilingual human would, but it can detect regularities in their vocalizations, movements and expressionsto relate them to specific contexts and, with sufficient information, propose probable meanings. It is a kind of algorithmic Rosetta Stone applied to the animal kingdom.
At the same time, the scientists themselves warn that this enthusiasm must be accompanied by caution: Translating is not just converting sounds into wordsIt also involves grasping tone, irony, intention, and context, something that is complicated even among humans. This difficulty is multiplied when dealing with other species.
Earth Species Project: the great experiment to decipher animal language
Among the most ambitious initiatives is the Earth Species Project (ESP), a non-profit organization born in Silicon Valley with an objective that, on paper, sounds almost impossible: decipher communication systems of as many species as possible on the planet and create AI tools capable not only of interpreting them, but also of generating messages that are understandable to those animals.
ESP combines engineers, machine learning experts, and animal behavior specialists. Their approach involves collecting huge volumes of sound recordings, videos, and environmental data associated with very specific contexts: feeding, play, reproduction, danger, socialization… With this information they train models that look for “minimal units of meaning” in the songs, chirps, clicks or roars of each species.
This project gained media visibility in 2023, when marine biologist Michelle Fournet, collaborating with ESP ecosystem technologies, succeeded interpret a type of humpback whale call in a context of social interaction. Although it wasn't a "word-for-word translation," it was proof that AI can help unravel previously undetectable communication patterns.
The people in charge of the Earth Species Project are working with ambitious time horizons: in about three decades they aspire to to be able to interpret, at least in part, the vocalizations of a great diversity of animalsThey even speak of a future in which nature documentaries will include subtitles automatically generated from the actual sounds of the filmed species, reducing the need for human narrators.
To support this leap, ESP has recently received around $17 million in fundingThis will allow them to hire more engineers, double the size of the current team, and accelerate the development of models like NatureLM-audio, which specializes in bioacoustics and is designed to learn directly from signals in nature.
NatureLM-audio and the “Rosetta Stone” of vocalizations
One of the most striking advances linked to this line of work is NatureLM-audio, an AI model that is trained with massive recordings of wildlife soundsIts main function, for now, is not to translate complete “phrases”, but to identify who emits the sound, in what context and with what systematic variations.
In tests conducted with different species, NatureLM-audio has achieved to distinguish with reasonable accuracy the sex, approximate age and number of individuals present in a recording. This, which may seem like a technical detail, is key for conservation projects: it allows us to know, for example, if a population of marine mammals is made up of many juveniles, if there are breeding females, or if an unusual movement is taking place.
In the case of endangered animals such as belugas or some cetaceans, researchers believe that systematically listen to their “calls” and understand them better It could be used to prevent collisions with ships, reduce noise in sensitive areas, or anticipate changes in migration routes before they translate into strandings or high mortality.
This “listen to conserve” approach marks a paradigm shift: instead of studying animals only from the outside, the aim is to… integrate their own communication as a priority data source to make decisions regarding environmental management and protection.
Elephants that call each other by name and other surprising discoveries
One of the most curious discoveries of recent years comes from the study of African elephants. By massively analyzing their vocalizations, scientists detected that, in certain situations, a specific sound was always directed at a single individual., One group memberswho was the one who answered. When that sound changed, a different elephant would reply.
Everything suggests that these colossi might use a form of "proper names", something like individual sound labels to address one another. These differences are so subtle that the human ear can barely distinguish them, but algorithms are able to see systematic patterns that suggest a more sophisticated linguistic structure than previously thought.
At sea, projects like CETI (which plays on the parallel with SETI, the program searching for extraterrestrial intelligence) focus on sperm whales. Their mission is to break down the clicks and rhythmic patterns emitted by these animals, relate them to observed situations and, with time and data, infer their approximate meaning.
If CETI manages to move forward in this direction, it would pave the way to decoding the languages of other marine mammalsHowever, each species would require its own specific strategy and models. There is no single "universal translator" that works for the entire animal kingdom, at least not yet.
Closer to the surface and to our everyday lives, remarkable progress has been made with farm animals: pigs, chickens and primates in zoos They are being monitored with high-resolution audio and video equipment to train AI to recognize moods, stress levels, or discomfort.
Chickens, pigs and monkeys: AI enters farms and zoos
In Canada, a group of researchers in Halifax is working with flocks of chickens whose clucking is analyzed using spectrograms and classification models. The idea is simple but powerful: to link each type of vocalization to an emotional state and a specific contextsuch as feeling calm, stressed, sick, or uncomfortable due to overcrowding.
If AI can reliably recognize when a flock of birds is suffering, farmers could Adjust ventilation, density, power supply, or lighting. long before obvious external signs appear, reducing suffering and improving productivity without waiting to see absences or injuries.
In Europe, other teams are using neural networks to analyze the vocal and gestural repertoire of pigs and various types of primates in zoos. Thanks to this, algorithms can classify vocalizations according to categories such as play, fear, aggression, or calmproviding caregivers with an “early warning” in case of social conflict or chronic stress.
These advances suggest that, in the not too distant future, many livestock farms could incorporate automatic listening and analysis systems that alert to animal welfare problems 24/7, without waiting for the periodic visit of the veterinarian or for mortality to skyrocket.
It's not just about producing more, but about to take seriously what animals have been telling us for a long time with their sounds and behaviors, now amplified and organized by AI tools.
Dogs and cats: when AI sneaks into the home
If there are animals that humans have dreamed of talking to for generations, it's dogs and cats. Daily coexistence reinforces the impression that “They understand us” much more than they can expressAI is starting to put numbers to that intuition.
At the University of Michigan, in collaboration with INAOE in Mexico, a team adapted the well-known Wav2Vec2 model—originally designed for human voice recognition—to the canine world. They trained the AI with recordings of 74 dogs in controlled emotional situations (joy, fear, frustration, play…) and asked him to classify the vocalizations.
The system was able to identify, with near-perfect accuracy 70 %the animal's emotional state, as well as variables such as breed, sex, and age range. This is the first time a model designed for human language has been successfully applied to barking, demonstrating that There is enough structure in those sounds to be interpreted algorithmically.
With cats, a research project at the University of Milan recorded meows in different everyday contexts—waiting for food, during brushing, in moments of isolation—and trained an algorithm to recognize the specific situation associated with each type of meowThe results showed an impressive 96% accuracy rate.
This type of research led to the development of MeowTalk, an application launched in 2020 by Javier Sánchez, a former Amazon Alexa engineer. The system has already processed more than one billion meows and is able to classify up to nine common communicative intentions, such as "I want to play", "I'm hungry" or "leave me alone", also customizing the profile for each cat.
Faces that speak: AI reading animal expressions
It's not all about the voice. Increasingly, researchers are combining audio with computer vision to analyze facial microexpressions and body postures in different species. In cats, for example, the so-called Feline Grimace Scale has become popular, which measures subtle changes in ears, eyes, whiskers and muzzle to assess whether the animal is suffering pain.
Apps like Tably use AI to capture and process images of the feline face and issue a pain probability score. In Japan, the CatsMe! system achieved over 95% accuracy in automatically detecting suffering in 2024, after being trained on thousands of photographs labeled by veterinarians.
Meanwhile, comparative studies have shown that humans and dogs share up to a 38% of recognizable facial expressionsThis figure rises to around 34% for cats. This suggests that there is common ground on which AI can operate to translate gestures into basic emotional states with increasing reliability.
The practical impact is enormous: veterinary clinics, shelters and animal welfare organizations could prioritize care for those animals that the system detects as being in the most paineven when they appear to be fine at first glance. AI thus acts as an “emotional magnifying glass” that complements—never replaces—professional judgment.
Beyond health, this automatic reading of expressions opens the door to toys, home cameras and smart necklaces that inform the guardian about whether their pet is bored, anxious or especially active, helping to adjust routines and enrich the environment.
Translators, talking avatars, and apps to talk to your pet
As research progresses, commercial products begin to appear that, relying on these scientific developments, offer approximate translations or playful visualizations of what dogs and cats “would say”Here the objective is no longer just scientific, but also entertainment and improving the human-animal bond.
The Earth Species Project itself is working on an application for AI-based translation for dogs and catswhich would work similarly to a chatbot: the user enters or records signals (barks, meows, movements), the system analyzes them in relation to its database and generates a textual response that represents the most likely intention.
In parallel, solutions such as Petpuls (collar that analyzes canine emotions)FluentPet-type button platforms—which allow the animal to press icons with pre-recorded words—or projects like Zoolingua, aimed at interpreting sounds, gestures and expressions in a multimodal way.
Another niche that has gained popularity is generators. talking pet videosThese tools allow you to upload a photo of a dog or cat, enter some text, and let the AI synchronize its mouth and expression to make it look like the animal is speaking. Although it's a purely aesthetic function, it helps familiarize the general public with the idea that one day that lip movement could be associated with a message based on real data.
Many of these apps offer free plans with limited usage and subscription models to access advanced features or unlimited generation, reflecting an emerging market around the “emotional translation” of pets.
From the farm to the veterinary hospital: practical applications and precautions
Beyond entertainment, artificial intelligence applied to animal language shows promise. to profoundly transform veterinary practice and the management of domestic and wild animals.Detecting pain, fear, or anxiety early can make the difference between timely treatment and an irreversible situation.
In clinical settings, audio and video analytics systems could be integrated into waiting rooms, operating theaters, or hospital wards to monitor changes in vocalizations and expressions that go unnoticed by the human eye, triggering alerts when something doesn't fit with a state of comfort.
In the natural environment, the same technologies would serve park rangers and conservationists to detect stress patterns in populations of elephants, rhinoceroses or seabirdsOptimize surveillance routes and anticipate conflicts with human activities such as navigation, agriculture, or construction.
However, professional organizations such as the Canadian Veterinary Medical Association insist on the need for a great caution in the use of these toolsThey emphasize that AI is still in its infancy, has biases and errors, and cannot in any case replace clinical examination, experience, and professional judgment.
The current recommendation is clear: use AI as supplementary supportnever as a substitute, and to be properly trained in its use. It is likely that veterinarians of the future will include subjects in data analysis and algorithm management in their training, becoming a kind of modern-day "Dolittle doctors," but always with their feet firmly on the scientific ground.
Ethical and social impact of being able to “listen” to animals
If we reach a point where AI allows animals to clearly express their preferences, their pain, or their rejection of certain practices, the The ethical implications will be enormousWe can no longer pretend that we don't know what they feel about certain forms of exploitation, transport, or confinement.
In the livestock sector, for example, clearer communication could force rethink cage sizes, population densities, or slaughter methodsIn the entertainment industry, circuses or shows with animals would see their legitimacy questioned even more if high levels of stress were demonstrated, reported directly by the performers themselves.
There are also risks of misuse of these technologies. The same system that would allow warn the whales to move away from a stranding area This could be used by unscrupulous whalers to lure them into capture or hunting areas. That is why many experts are calling for clear regulatory frameworks and strict ethical codes.
Another delicate issue is that of anthropomorphism: the temptation to to project human emotions and thoughts onto animal signals which may not mean exactly what we think. Here, AI can exacerbate the problem if it translates messages into natural language that sounds too "human," creating the illusion of a dialogue that is actually a statistical model.
Even so, most research agrees that giving "voice" to other species can strengthen our empathy towards them and towards ecosystemshelping in the fight against climate change and the loss of biodiversity through a more horizontal relationship with the rest of living beings.
Everything suggests that, over time, we will coexist with AI systems that integrate audio, video, and physiological data into multimodal models, capable of producing bidirectional translations adapted to the "vocabulary" of each individual, not replacing our direct relationship, but enriching it.
The real possibility that an algorithm can tell us with reasonable reliability what a dog, a cat, an elephant, or a whale is trying to convey will change the way we see them, care for them, and decide what is acceptable to do with them, bringing us a little closer to that dream world in which all creatures finally have something to tell us and we, for the first time, have serious tools to truly listen to them.
