Google Unveils DolphinGemma AI to Decode Dolphin Communication

The groundbreaking model, trained on decades of research, is now operational on Pixel phones with plans for an open-source release this summer.

DolphinGemma, developed by Google in collaboration with the Wild Dolphin Project and Georgia Tech, processes dolphin vocalizations to identify patterns and potential meanings.
The AI model, built on a 400M parameter architecture, runs in real-time on Pixel phones, enabling field researchers to analyze dolphin sounds directly in the ocean.
Decades of data from the Wild Dolphin Project, the longest-running underwater dolphin research initiative, provided the foundation for training the model.
A companion system, CHAT, is under development to facilitate two-way interactions by associating synthetic whistles with objects dolphins enjoy, like seagrass and scarves.
Google plans to release DolphinGemma as an open-source model this summer, aiming to expand its application to other cetacean species and accelerate global research.