Hawaiian monk seals communicate.
For a long time, these rare animals were considered almost voiceless underwater. In 2025, scientists first heard the true diversity of their voices. And in 2026, artificial intelligence learned to find these signals among thousands of hours of ocean sound.
The ocean was not silent. We just did not yet know how to listen to it.
A voice that was heard but not listened to
The Hawaiian monk seal is endemic to the Hawaiian Archipelago: in the wild, this species is found nowhere else on the planet. Its population numbers about 1 600 animals, so every new method of observation can be important for the conservation of the species.
Studying seals is not easy. They spend most of their lives in the water, and their habitats are scattered across remote islands and atolls. Visual observations provide only isolated fragments of what is happening.
But sound remains in the ocean even when the animal itself is not visible.
The first major change occurred on 12 November 2025, when the journal Royal Society Open Science published a field study of underwater vocalizations of free-ranging Hawaiian monk seals.
Scientists placed hydrophones at five sites—from the main Hawaiian Islands to the remote northwestern atolls. In more than 4 500 hours of recordings they found over 23 000 vocalizations.
Instead of the few signals known previously, the researchers described 25 types of vocal calls, including 20 previously unknown.
Nineteen new types turned out to be combinatorial: the animals combined several sounds into more complex sequences. Most signals sounded at frequencies below one kilohertz, lasted from fractions of a second to seven seconds, and were often repeated in series.
Seals, which had been considered almost voiceless, turned out to be surprisingly talkative.
From the discovery of voices to automatic listening
The next step was taken 24 August 2026.
In the journal Marine Mammal Science a new peer-reviewed study was published Description, Detection, and Classification of Hawaiian Monk Seal Vocalizations Through Deep-Learning Methods: Biological and Conservation Applications.
If the 2025 study showed the richness of the seals' acoustic world, the new research proposed a tool capable of automatically finding their voices in vast arrays of recordings.
To train the neural network, the researchers used 4 989 manually annotated signals, obtained from two adult seals kept separately under controlled conditions.
Each sound was represented as a spectrogram—an image in which frequency, duration, and intensity become a discernible pattern.
The scientists fine-tuned the YOLO computer vision system. Typically, such algorithms recognize objects in photos and videos, but here the neural network searched for voices in acoustic images.
The model achieved 87% weighted accuracy and detected 48 324 signals in 16 262 hours of recordings of the same animals.
What would have required months of continuous listening by a human, the system could search for automatically.
Trial by the living ocean
A clean recording and the real underwater environment are completely different worlds.
In the ocean, a seal's voice mixes with waves, currents, sounds of other animals, ship movement, and constant background noise. To test the model under more challenging conditions, the scientists combined confirmed signals with recordings of real marine sounds at different signal-to-noise ratios.
After retraining, the algorithm detected 2 049 vocalizations in 12 hours of field recordings, made in important seal habitats.
The periods of acoustic presence marked by the neural network coincided with the results of manual annotation.
This is not yet a universal translator from seal language. But it is already a working tool capable of hearing the animal within the noisy ocean.
Why conservation needs the voice
A hydrophone can remain underwater much longer than a researcher or research vessel. It continues to listen at night, in bad weather, and when the animal cannot be seen from the surface.
Passive acoustic monitoring can help scientists determine:
- where and when seals appear;
- how their vocal activity changes;
- which areas are especially important for breeding;
- how animals use remote habitats;
- how human noise affects their ability to hear each other.
Artificial intelligence does not replace a biologist. It finds important episodes in a huge sound archive, and a person checks the result, studies the context, and tries to understand the meaning of what is heard.
Thus, ocean acoustics becomes not only a way of knowing, but also a form of protection.
When the ocean gets a voice
We are used to imagining the ocean through images: blue depths, coral reefs, silhouettes of animals in the water.
But the ocean exists not only in light. It exists in sound.
Voices help marine animals find each other where visibility disappears. They accompany meetings, rivalry, breeding, foraging, and care for offspring.
In 2025, scientists discovered the richness of voices of the Hawaiian monk seal. In 2026, artificial intelligence learned to recognize them among thousands of hours of ocean sound.
It did not translate the seals' speech into human language.
It did something more cautious and perhaps more important: it helped humans stop mistaking inaudibility for silence.
They were considered voiceless. But the ocean always knew: they have a voice.
