For decades, scientists have dreamed of not just listening to dolphins but actually understanding them. Now, thanks to Google's new AI project, that dream might be closer than ever.
Today, on National Dolphin Day, Google, in collaboration with Georgia Tech and the Wild Dolphin Project (WDP), has announced DolphinGemma — a powerful AI model that can study and generate dolphin-like sounds. This breakthrough moves us one step closer to real interspecies communication and opens new possibilities for connecting with the marine world.
Since 1985, the Wild Dolphin Project has been studying wild Atlantic spotted dolphins (Stenella frontalis) in the Bahamas. This is the longest underwater dolphin research project in the world. Their approach is special: they study dolphins "In Their World, on Their Terms," without disturbing them.
Over the years, WDP has collected a huge amount of underwater videos and audio, matched with the identities and life stories of individual dolphins. Their focus is on understanding natural dolphin communication. Some examples include:
Knowing who is making the sound and why is key to understanding dolphin language. This deep research is what made DolphinGemma possible.
DolphinGemma is Google's new AI model trained on decades of dolphin sound data from the WDP. It uses cutting-edge Google audio technologies like SoundStream to efficiently process complex dolphin sounds.
This AI model has around 400 million parameters and can even run directly on Google Pixel phones used by researchers in the field.
Early tests show that DolphinGemma can generate realistic whistles and burst pulses similar to natural dolphin sounds. Researchers are excited to use this model to find hidden patterns in dolphin communication much faster than before.
Beyond analyzing dolphin sounds, the WDP is also working on two-way communication with dolphins using a system called CHAT (Cetacean Hearing Augmentation Telemetry).
Here’s how it works:
The latest version of CHAT, powered by the Google Pixel 9, runs advanced AI models in real time, making interaction faster and more accurate.
This summer, Google plans to release DolphinGemma as an open-source model. While it’s trained on Atlantic spotted dolphin sounds, it can also help researchers working with bottlenose or spinner dolphins.
By giving scientists access to this technology, Google hopes to speed up discoveries about dolphin communication around the world.
The combination of long-term dolphin research, Georgia Tech engineering, and Google's AI expertise is creating a future where humans and dolphins might finally have meaningful conversations. Thanks to DolphinGemma, we’re no longer just listening we’re starting to understand.
The ocean still holds many secrets, but today, we’re a little closer to uncovering them.
DolphinGemma analyzes patterns in dolphin sounds to predict and generate realistic dolphin-like sounds, helping researchers uncover the structure of dolphin communication.
CHAT (Cetacean Hearing Augmentation Telemetry) is a system designed to create two-way interaction between dolphins and researchers by associating synthetic whistles with objects dolphins like.
Yes, Google plans to release DolphinGemma as an open model this summer, allowing researchers worldwide to adapt it for studying various dolphin species.
Please login to post a comment.
No comments yet. Be the first to comment!