Machine learning is uncovering patterns in whale clicks, dolphin whistles, bird songs and elephant calls that humans struggle to detect, but understanding animals is still very different from translating them.
Artificial intelligence is giving scientists a new way to investigate one of biology’s oldest questions: what are animals communicating to one another?
AI can now process enormous collections of animal sounds, compare subtle acoustic differences and identify recurring patterns far faster than researchers working manually. From sperm whale clicks in the Caribbean to elephant rumbles in Africa, machine-learning systems are revealing previously hidden structure in communication across species.
The results are significant, but they require careful interpretation. Scientists have not developed a universal animal translator, nor demonstrated that animal communication operates like human language. Instead, AI is becoming an increasingly powerful analytical tool, helping researchers determine which sounds occur, who produces them, when they appear and what behaviours accompany them.
That distinction is central to understanding the emerging field.
AI Finds Structure in Animal Communication
Sperm whales provide one of the clearest examples of what machine learning can contribute.
The whales communicate using patterned sequences of clicks known as codas. Project CETI, an interdisciplinary research initiative focused on sperm whales around Dominica in the eastern Caribbean, is combining extensive acoustic recordings, behavioural observations, robotics and machine learning to investigate how those vocalizations are structured.
A 2024 study published in Nature Communications analyzed sperm whale vocalizations and identified previously undescribed features researchers called “rubato” and “ornamentation.” Combined with rhythm and tempo, these features create a much larger inventory of distinguishable codas than simple classifications had suggested.
Researchers concluded that sperm whale vocalizations contain contextual and combinatorial structure. That does not mean individual clicks can yet be translated into human sentences, but it provides evidence that the communication system is considerably richer than previously understood.
Dolphins are another major target.
The Wild Dolphin Project has studied a community of wild Atlantic spotted dolphins in the Bahamas since 1985, accumulating decades of recordings paired with observations of individual dolphins and their behaviour. In 2025, Google introduced DolphinGemma in collaboration with the project and researchers at Georgia Tech.
The roughly 400-million-parameter AI model was trained on the project’s acoustic database to process dolphin sounds, recognize patterns and predict likely subsequent sounds in a sequence.
Researchers hope that finding recurring clusters and sequences will help expose structure within dolphin communication. A separate experimental system called CHAT explores whether humans and dolphins can establish a limited shared vocabulary by associating artificial whistles with objects.
Neither system means researchers can simply ask a dolphin a question and receive a translated answer. They are experimental approaches for testing relationships between sound, behaviour and meaning.
Elephants have produced an especially intriguing result.
A 2024 study in Nature Ecology & Evolution used machine learning to examine calls from wild African elephants. The model found acoustic information associated with the intended receiver of a call. Researchers then conducted playback experiments and found that elephants responded differently to calls originally directed toward them than to calls directed toward another elephant.
The evidence suggests elephants use individually specific, name-like vocal labels, probably without simply imitating the recipient’s own calls. Researchers have been careful to describe them as name-like rather than claiming they are identical to human names.
From Bird Songs to Vast Acoustic Archives
Bird research illustrates another strength of AI: scale.
Modern acoustic recorders can collect thousands of hours of environmental sound. Listening to all of it manually is impractical. Systems such as BirdNET use deep learning to identify species from recordings, turning enormous sound archives into searchable ecological information.
BirdNET, a collaboration involving the Cornell Lab of Ornithology and Chemnitz University of Technology, says its current tools can recognize more than 6,000 species. The technology converts recordings into spectrograms and uses neural networks to identify characteristic acoustic patterns.
This type of AI is primarily identifying who is calling rather than determining what the animal is saying. But the same ability to process huge datasets creates opportunities to investigate how vocalizations change with location, season, social circumstances and behaviour.
Broader foundation models are also emerging. Earth Species Project reported that its NatureLM-audio model, designed specifically for animal sounds, was open-sourced and accepted at the International Conference on Learning Representations in 2025. The organization is developing tools intended to work across diverse species rather than concentrating exclusively on one animal.
The underlying strategy resembles other applications of machine learning: provide algorithms with enormous amounts of data and allow them to detect statistical relationships that might escape human observers.
But animal communication creates an unusual problem. Human language models can be trained on text whose meaning humans already understand. There is no equivalent dictionary explaining what a sperm whale click or elephant rumble means.
Researchers therefore need more than audio. They need observations identifying the caller and receiver, their behaviour, environmental conditions and what happens before and after a sound. Playback experiments can then test whether animals respond as predicted.
AI may eventually help scientists connect increasingly complex vocal patterns with specific behaviours, individuals or circumstances. It could also strengthen conservation by making it easier to monitor species and understand how human-generated noise affects their interactions.
The emerging evidence nevertheless supports a more measured idea than the prospect of an instant animal translator.
AI is allowing humans to listen differently.
For the first time, researchers can systematically examine acoustic datasets at a scale and level of detail that would have been extraordinarily difficult using human analysis alone. What those patterns ultimately reveal about the minds and social worlds of other species remains one of the most compelling unanswered questions in modern biology.
About the Author Brad Socha is the founder of The Universal Record, focused on sourced, factual global reporting. Coverage includes international news, geopolitics, technology, and major developments.
