Google’s DolphinGemma AI Cracks the Code of Dolphin Language

As part of National Dolphin Day, Google unveiled DolphinGemma—a revolutionary AI model specifically created to translate and mimic the vocalizations of dolphins. Developed through collaboration with Georgia Institute of Technology and Wild Dolphin Project (WDP), DolphinGemma is a promising development toward unraveling the sophisticated "language" of Atlantic spotted dolphins.

WDP, which operates the world's longest continuous underwater dolphin research study in the Bahamas since 1985, has recorded a vast library of underwater sounds and images. They provide recordings, tagged to particular activities, allowing scientists to link vocalizations such as signature whistles, clicks, and squawks with social activities such as mating ceremonies, conflicts, and bonding experiences.

DolphinGemma leverages this treasure chest with Google's SoundStream and a powerful 400-million-parameter model to examine and create dolphin-like sounds. Its ability to predict gives it greater insight into vocal structure and patterns, allowing it to better understand how dolphins may be employing sound as a means of communication.

Optimized for mobile, the AI can even be executed on Google Pixel smartphones, making marine biologists capable of carrying out spot field studies. Further, the tool will make current systems such as CHAT (Cetacean Hearing Augmentation Telemetry) better, where particular sounds are linked with dolphin-preferred objects, facilitating more interactive and responsive communication situations.

Also Read: Samsung Suspends One UI 7 Update After Bug Locks Users Out

Looking into the future, Google intends to open-source DolphinGemma around summer 2025, allowing marine researchers around the world to modify and implement its capabilities to other dolphin populations. Through this innovation, humans' aspiration of having a unified communication platform for humans and dolphins moves closer to becoming a reality.