Google, in collaboration with Georgia Tech and the Wild Dolphin Project (WDP), has developed a groundbreaking artificial intelligence model called DolphinGemma, poised to revolutionize the quest to understand dolphin communication. For decades, scientists have sought to unravel the mysteries of dolphin language—those intricate whistles, clicks, and burst pulses. Now, DolphinGemma uses advanced pattern recognition to analyze and even predict dolphin vocalizations, much like AI models developed for human languages.
Built upon Google’s open-source Gemma models and powered by the innovative SoundStream audio technology, DolphinGemma efficiently processes complex dolphin sounds for analysis. The lightweight 400 million-parameter model was specifically designed for field use—capable of running directly on mobile devices like Google Pixel phones, which researchers use during their expeditions.
Trained with WDP’s exhaustive acoustic archives collected over decades, DolphinGemma can analyze thousands of hours of dolphin vocalizations alongside detailed behavioral annotations—something previously infeasible for human researchers. Scientists anticipate that the AI will uncover communication patterns such as individualized dolphin whistles used as “names” and unique sound sequences used during social interactions, potentially revealing a true dolphin “vocabulary.”
Alongside DolphinGemma, the collaborative development with Georgia Tech has produced the CHAT system—a portable, real-time technology leveraging Pixel smartphones to enable simple, direct communication attempts with dolphins. Designed to establish a shared vocabulary through synthetic whistles associated with specific objects, CHAT combined with AI advances is set to make human-dolphin interaction more fluid and responsive than ever before.
Google will release DolphinGemma as an open-source model this summer. While currently trained on Atlantic spotted dolphins, the system can potentially be adapted for other dolphin and whale species. Experts believe this breakthrough brings us closer than ever to meaningful interspecies communication, opening a new chapter in animal cognition research and our relationship with intelligent marine life.