Google has unveiled a groundbreaking AI model named DolphinGemma, designed specifically to understand and communicate with dolphins. Announced in April 2025, this innovative large language model (LLM) represents a major leap forward in decoding the complex vocalizations of dolphins and potentially establishing two-way communication between humans and these intelligent marine mammals.
DolphinGemma was developed in collaboration with the Wild Dolphin Project (WDP), a nonprofit organization that has conducted the world’s longest-running underwater dolphin research since 1985. The model was trained on over 40 years of meticulously collected underwater audio-visual data of Atlantic spotted dolphins in the Bahamas. By leveraging Google’s advanced SoundStream audio processing technology, DolphinGemma converts dolphin sounds into discrete audio tokens and predicts subsequent vocalizations, much like how human language models anticipate the next word in a sentence.
With approximately 400 million parameters, DolphinGemma is optimized for efficiency and can even run on Google Pixel smartphones, enabling researchers to perform real-time analysis and interaction in the field. This portability is crucial for capturing and interpreting dolphin communication in their natural habitat.
One of the most exciting aspects of DolphinGemma is its ability not only to decode natural dolphin vocal patterns—such as signature whistles used for individual identification, burst-pulse sounds during social conflicts, and clicks associated with mating or hunting—but also to generate realistic dolphin-like sound sequences. This capability opens the door to interactive communication, where humans might respond to dolphins using synthesized sounds that the animals can recognize and mimic.
Google and the Georgia Institute of Technology have further developed the CHAT (Cetacean Hearing Augmentation Telemetry) system, an underwater computer that works alongside DolphinGemma to establish a shared vocabulary. By associating synthetic whistle sounds with specific objects dolphins enjoy, researchers encourage dolphins to mimic these sounds to request items, effectively creating a two-way communication channel. As DolphinGemma continues to decode more natural dolphin sounds, this shared vocabulary can expand, paving the way for deeper interspecies interaction.
Google plans to open-source DolphinGemma by summer 2025, inviting the global research community to contribute to and expand this technology. The open-source release is expected to accelerate studies across different dolphin species and foster broader applications in marine biology, animal behavior, and conservation.
In summary, DolphinGemma marks a historic advancement in AI-driven animal communication research. By combining cutting-edge machine learning with decades of biological data, Google is bridging the gap between humans and one of the ocean’s most intelligent species. This breakthrough not only deepens our understanding of dolphin society but also sets the stage for a future where meaningful dialogue between species may become a reality.
